1. An intelligent medical big data processing method based on artificial intelligence is applied to an intelligent medical system, the intelligent medical system is in communication connection with a plurality of intelligent medical subscription devices, and the method comprises the following steps:

for any target subscription medical resource of the intelligent medical subscription equipment, acquiring medical resource behavior big data of the target subscription medical resource;

processing the medical resource behavior big data to obtain search intention information distribution of the medical resource behavior big data, wherein the search intention information distribution comprises intention distribution information of a search pointing target of a target medical pushing tag in the medical resource behavior big data;

and determining big data mining information of the medical resource behavior big data according to the search intention information distribution, wherein the big data mining information comprises mining label information of a search pointing target of the target medical push label in the medical resource behavior big data.

2. The intelligent medical big data processing method based on artificial intelligence as claimed in claim 1, wherein the processing the medical resource behavior big data to obtain the search intention information distribution of the medical resource behavior big data comprises:

and calling a search intention classification unit to process the medical resource behavior big data so as to obtain the search intention information distribution of the medical resource behavior big data, wherein the search intention classification unit is obtained by performing network weight learning and updating on the reference medical resource behavior data of the target subscription medical resource.

3. The intelligent medical big data processing method based on artificial intelligence as claimed in claim 2, wherein before the step of calling the search intention classification unit to process the medical resource behavior big data, the method further comprises:

acquiring a reference data sequence, wherein the reference data sequence comprises first reference medical resource behavior data of an original subscribed medical resource and second reference medical resource behavior data of a target subscribed medical resource;

and performing network weight learning updating on the second reference medical resource behavior data and learning and training the first reference medical resource behavior data to obtain a reference machine learning unit, wherein when the trained reference machine learning unit meets a training termination requirement, the trained reference machine learning unit is used as a search intention classification unit, and the search intention classification unit is used for identifying intention distribution information of a search pointing target of the target medical push label in the medical resource behavior data of the target subscription medical resource.

4. The intelligent medical big data processing method based on artificial intelligence as claimed in claim 3, wherein the reference data sequence further includes reference intention distribution information of search pointing to target of target medical push tag in the first reference medical resource behavior data;

the reference machine learning unit for training the first reference medical resource behavior data by performing network weight learning update on the second reference medical resource behavior data comprises:

calling the reference machine learning unit to perform feature extraction on the second reference medical resource behavior data to obtain a second decision intention distribution feature of the second reference medical resource behavior data;

unifying feature vectors of second shared intention component features or second continuous shared intention component features included in the second decision intention distribution features to obtain second shared intention component features or second continuous shared intention component features after vector unification, wherein the second shared intention component features or the second continuous shared intention component features after vector unification include a plurality of intention components, and each intention component corresponds to one behavior interest behavior in the second reference medical resource behavior data;

respectively calculating redundant exclusion information quantities for each intention component in the plurality of intention components, and obtaining a third decision evaluation index according to the redundant exclusion information quantities of all the intention components, the feature range and the feature level of the second decision intention distribution feature;

calculating a maximum square decision evaluation index for each intention component in the plurality of intention components, and obtaining a fourth decision evaluation index according to the maximum square decision evaluation indexes of all the intention components, the feature range of the second decision intention distribution feature and the feature level;

determining a first target decision evaluation index of the reference machine learning unit according to the third decision evaluation index and the fourth decision evaluation index;

calling the reference machine learning unit to perform feature extraction on the first reference medical resource behavior data to obtain a first decision intention distribution feature of the first reference medical resource behavior data;

determining a second target decision evaluation index of the reference machine learning unit according to the first decision intent distribution feature and the reference intent distribution information;

training the reference machine learning unit according to the first objective decision evaluation index and the second objective decision evaluation index.

5. The intelligent medical big data processing method based on artificial intelligence as claimed in claim 4, wherein the first decision intention distribution feature comprises a first sharing intention component feature and a first continuous sharing intention component feature, and the reference intention distribution information comprises a feature range, a feature level and an intention interest point of the target medical push tag;

the determining a second objective decision evaluation indicator of the reference machine learning unit according to the first decision intent distribution feature and the reference intent distribution information includes:

determining a first decision evaluation index according to the first sharing intention component characteristic, the intention interest point of the target medical pushing label, and the quantity of the first reference medical resource behavior data;

determining a second decision evaluation index according to the first continuous sharing intention component feature, the quantity of the first reference medical resource behavior data, the feature range and the feature level of the intention object;

determining a second target decision evaluation index of the reference machine learning unit according to the first decision evaluation index and the second decision evaluation index.

6. The intelligent medical big data processing method based on artificial intelligence according to claim 4, wherein the first objective decision evaluation index comprises a third decision evaluation index and a fourth decision evaluation index, and the second objective decision evaluation index comprises a first decision evaluation index and a second decision evaluation index;

the training the reference machine learning unit according to the first objective decision evaluation indicator and the second objective decision evaluation indicator comprises:

obtaining a first influence weight corresponding to the first decision evaluation index, a second influence weight corresponding to the second decision evaluation index, a third influence weight corresponding to the third decision evaluation index, and a fourth influence weight corresponding to the third decision evaluation index;

performing weight comprehensive calculation on the second target decision evaluation index and the first target decision evaluation index according to the first influence weight, the second influence weight, the third influence weight and the fourth influence weight to obtain a target decision evaluation index;

and adjusting the unit weight of the reference machine learning unit according to the target decision evaluation index.

7. The intelligent medical big data processing method based on artificial intelligence as claimed in claim 2, wherein the search intention classification unit comprises a search intention extraction node and a search intention fusion node;

the step of processing the medical resource behavior big data by calling the search intention classification unit to obtain the search intention information distribution of the medical resource behavior big data comprises the following steps:

calling the search intention extraction node to perform feature extraction on the medical resource behavior big data so as to obtain initial search intention features of the medical resource behavior big data;

calling the search intention fusion node to perform convolution processing and sampling processing on the initial search intention characteristic to obtain a first search intention characteristic;

calling the search intention fusion node to perform compression processing and excitation processing on the initial search intention characteristics to obtain influence weights corresponding to the initial search intention characteristics, and performing weighted calculation on the initial search intention characteristics according to the influence weights to obtain second search intention characteristics;

and performing fusion processing on the first search intention characteristic and the second search intention characteristic to obtain the search intention information distribution of the medical resource behavior big data.

8. The intelligent medical big data processing method based on artificial intelligence is characterized in that the search intention information distribution comprises a shared search intention feature and a continuously shared search intention feature, the shared search intention feature comprises confidence of each behavior interest behavior in the medical resource behavior big data to an intention interest point of a target for searching the target medical push label, and the continuously shared search intention feature comprises a feature range and feature level data corresponding to each behavior interest behavior in the medical resource behavior big data;

the big data mining information for determining the big data of the medical resource behaviors according to the search intention information distribution comprises the following steps:

determining an intention interest point of the target medical pushing label in the medical resource behavior big data, which is searched and pointed to by the target, according to the shared search intention characteristics;

determining an intention object of the target in the medical resource behavior big data pointed by the search of the target medical push label according to the intention interest point and the characteristic range and the characteristic level data corresponding to the behavior interest behavior at the intention interest point;

and taking the intention object of the target medical pushing label pointed by the search as the mining label information of the target medical pushing label pointed by the search.

9. The intelligent medical big data processing method based on artificial intelligence according to any one of claims 1-8, characterized in that the method further comprises:

acquiring E-commerce hotspot information of each relevant mining tag in a corresponding target medical E-commerce data source according to the big data mining information;

acquiring the E-commerce hotspot distribution characteristics of the relevant mining tags according to the E-commerce hotspot information of the relevant mining tags;

fusing the E-commerce hotspot distribution characteristics of the relevant mining labels to obtain mining hotspot distribution characteristics of the target medical E-commerce data source;

performing medical e-commerce push attribute decision on the target medical e-commerce data source according to the mining hotspot distribution characteristics to obtain at least one medical e-commerce push attribute of the target medical e-commerce data source;

and carrying out corresponding medical e-commerce information pushing on the intelligent medical subscription equipment according to at least one medical e-commerce pushing attribute of the target medical e-commerce data source.

10. An intelligent medical system, comprising at least one storage medium and at least one processor, the at least one storage medium configured to store computer instructions; the at least one processor is configured to execute the computer instructions to perform the method for intelligent medical big data processing based on artificial intelligence according to any one of claims 1 to 9.

Background

With the rapid growth of networks and technologies in recent years, the generation speed and quantity of digital data have been rapidly increased, and thus, all industries have actively invested in large data applications so far, and the intelligent medical industry is no exception, so that it is expected to speed up the search intention demand of users, and further find out effective information service approaches, thereby providing convenient intelligent medical services for users.

At present, the conventional big data mining mainly performs AI training through reference medical resource behavior data of original subscribed medical resources and labels of the reference medical resource behavior data, then performs manual labeling on the reference medical resource behavior data of target subscribed medical resources, and optimizes a detection network model through the reference medical resource behavior data of the target subscribed medical resources with the labels, so that the aim of detecting mining labels of the medical resource behavior data of the target subscribed medical resources by the AI can be fulfilled. The manual intervention wastes manpower and consumes time, and the manual intervention may cause inaccurate labeling of the reference medical resource behavior data due to subjective factors, so that the effect of processing the medical resource behavior data is poor.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present disclosure provides an intelligent medical big data processing method and an intelligent medical system based on artificial intelligence.

In a first aspect, the present disclosure provides an intelligent medical big data processing method based on artificial intelligence, applied to an intelligent medical system, where the intelligent medical system is in communication connection with a plurality of intelligent medical subscription devices, and the method includes:

for any target subscription medical resource of the intelligent medical subscription equipment, acquiring medical resource behavior big data of the target subscription medical resource;

processing the medical resource behavior big data to obtain search intention information distribution of the medical resource behavior big data, wherein the search intention information distribution comprises intention distribution information of a search pointing target of a target medical pushing tag in the medical resource behavior big data;

and determining big data mining information of the medical resource behavior big data according to the search intention information distribution, wherein the big data mining information comprises mining label information of a search pointing target of the target medical push label in the medical resource behavior big data.

In a second aspect, the disclosed embodiment further provides an artificial intelligence based intelligent medical big data processing system, where the artificial intelligence based intelligent medical big data processing system includes an intelligent medical system and a plurality of intelligent medical subscription devices in communication connection with the intelligent medical system;

the intelligent medical system is used for:

for any target subscription medical resource of the intelligent medical subscription equipment, acquiring medical resource behavior big data of the target subscription medical resource;

processing the medical resource behavior big data to obtain search intention information distribution of the medical resource behavior big data, wherein the search intention information distribution comprises intention distribution information of a search pointing target of a target medical pushing tag in the medical resource behavior big data;

and determining big data mining information of the medical resource behavior big data according to the search intention information distribution, wherein the big data mining information comprises mining label information of a search pointing target of the target medical push label in the medical resource behavior big data.

According to any one of the above aspects, in the embodiments provided by the present disclosure, the medical resource behavior big data may be processed to obtain the search intention information distribution of the medical resource behavior big data, where the search intention information distribution includes intention distribution information of a search target of a target medical push tag in the medical resource behavior big data. Then, according to the search intention information distribution of the medical resource behavior big data, the big data mining information of the medical resource behavior big data can be determined. Compared with the method for manually marking the reference medical resource behavior data of the target subscription medical resource, the method for manually marking the reference medical resource behavior data of the target subscription medical resource is not needed, and the big data mining information of the big data of the medical resource behavior can be automatically mined, so that the efficiency and the precision of big data mining are improved.

Drawings

FIG. 1 is a schematic diagram of an application scenario of an intelligent medical big data processing system based on artificial intelligence according to an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart illustrating a method for processing intelligent medical big data based on artificial intelligence according to an embodiment of the disclosure;

FIG. 3 is a functional block diagram of an intelligent medical big data processing device based on artificial intelligence according to an embodiment of the disclosure;

fig. 4 is a schematic block diagram of an intelligent medical system for implementing the above intelligent medical big data processing method based on artificial intelligence according to an embodiment of the present disclosure.

Detailed Description

The following describes in detail aspects of embodiments of the present disclosure with reference to the drawings attached hereto. In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular system structures, interfaces, techniques, etc. in order to provide a thorough understanding of the particular embodiments of the disclosure.

Fig. 1 is a schematic view of a smart medical big data processing system 10 based on artificial intelligence according to an embodiment of the present disclosure. The intelligent medical big data processing system 10 based on artificial intelligence can comprise an intelligent medical system 100 and an intelligent medical subscription device 200 which is in communication connection with the intelligent medical system 100. The intelligent medical big data processing system 10 based on artificial intelligence shown in fig. 1 is only one possible example, and in other possible embodiments, the intelligent medical big data processing system 10 based on artificial intelligence may also only include at least part of the components shown in fig. 1 or may also include other components.

In this embodiment, the intelligent medical system 100 and the intelligent medical subscription device 200 in the intelligent medical big data processing system 10 based on artificial intelligence can cooperate to execute the intelligent medical big data processing method based on artificial intelligence described in the following method embodiments, and the detailed description of the following method embodiments can be referred to in the execution steps of the intelligent medical system 100 and the intelligent medical subscription device 200.

Before describing the embodiments of the present disclosure, a scenario of first reference medical resource behavior data under an originally subscribed medical resource is first described below. In the present disclosure, a batch of reference data sets may be selected for artificial intelligence learning, that is, a reference medical resource behavior data set for artificial intelligence learning is selected, where the reference medical resource behavior data set includes first reference medical resource behavior data under an original subscription medical resource and second reference medical resource behavior data under a target subscription medical resource. The first reference medical resource behavior data under the original subscribed medical resource means: and the data sample of the original subscribed medical resource comprises the reference medical resource behavior data and the label data with the target label (which can be marked as reference intention distribution information). In addition, the reference medical resource behavior data under the originally subscribed medical resource is the simulated medical resource behavior data, that is, the medical resource behavior data can be automatically simulated. The second reference medical resource behavior data under the target subscription medical resource means: and subscribing data samples of the medical resources by the targets in the new medical service version, namely medical resource behavior data in the actual application environment. And the reference medical resource behavior data under the target subscription medical resource only includes the medical resource behavior data and is not labeled, that is, the target subscription medical resource is a scene concerned by the subscription medical resource migration task.

In one embodiment, the reference machine learning unit may be trained according to first reference medical resource behavior data of an original subscribed medical resource, reference intention distribution information of a search pointing target of a target medical push tag in the first reference medical resource behavior data, and second reference medical resource behavior data of a target subscribed medical resource.

Taking a specific service scenario as an example: the original subscribed medical resources are the simulated medical resource behavior data and the reference intention distribution information of the medical consultation search object, and the reference intention distribution information comprises at least one type. The target subscribed medical resources are medical resource behavior data in the actual application environment. Of course, in a specific service scenario, the number of targets of the search pointing target referring to the intention distribution information may be adaptively adjusted according to the service scenario or the service requirement. It should be noted that the number of targets included in the reference intention distribution information in the first reference medical resource behavior data is greater than or equal to the number of targets of search direction targets included in the finally identified medical resource behavior big data.

In one embodiment, the smart medical system 100 trains the reference machine learning unit according to the first reference medical resource behavior data of the originally subscribed medical resource, the reference intention distribution information of the target medical push tag in the first reference medical resource behavior data, and the second reference medical resource behavior data of the target subscribed medical resource. For example, the reference machine learning unit comprises a search intention extraction node and a search intention fusion node, and each batch of training is performed by inputting first reference medical resource behavior data of an original subscribed medical resource and second reference medical resource behavior data of a target subscribed medical resource into the reference machine learning unit together, wherein the first reference medical resource behavior data comprises reference intention distribution information of the target medical push label, search of the target medical push label points to the target, the target medical push label is of at least one type, and the second reference medical resource behavior data does not carry the reference intention distribution information. Feature extraction is respectively carried out on the first reference medical resource behavior data and the second reference medical resource behavior data through the search intention extraction nodes, and search intention information distribution corresponding to the first reference medical resource behavior data and search intention information distribution corresponding to the second reference medical resource behavior data are respectively obtained. Then, the reference machine learning unit is trained according to the search intention information distribution corresponding to the first reference medical resource behavior data, the redundancy exclusion information amount of the search intention information distribution corresponding to the second reference medical resource behavior data, and the reference intention distribution information of the search pointing target of the target medical push label of the first reference medical resource behavior data.

When the trained reference machine learning unit meets the training termination requirement, the reference machine learning unit is used as a search intention classification unit, and the search intention classification unit can be used for detecting a search intention pointing to a target of a target medical push label in medical resource behavior data of a target subscription medical resource. For example, the intelligent medical system 100 determines the big data mining information of the big data of the medical resource behavior by determining the search intention information distribution of the big data of the medical resource behavior, and determining the big data mining information of the big data of the medical resource behavior according to the search intention information distribution, wherein the big data mining information comprises mining tag information of a search pointing target of a target medical pushing tag in the big data of the medical resource behavior.

In this way, the machine learning network trained by the redundancy elimination information amount of the reference medical resource behavior data of the original subscribed medical resource with the reference intention distribution information and the target subscribed medical resource reference medical resource behavior data without the reference intention distribution information can be migrated to the task medical resource behavior data of the target subscribed medical resource without labeling the medical resource behavior data of the target subscribed medical resource, so that the labeling cost of the medical resource behavior data of the target subscribed medical resource is greatly saved, even avoided, and the intention classification detection of the specified object in the medical resource behavior data of the target subscribed medical resource is realized.

To solve the technical problem in the background art, fig. 2 is a schematic flow chart of an intelligent medical big data processing method based on artificial intelligence according to an embodiment of the disclosure, which can be executed by the intelligent medical system 100 shown in fig. 1, and the intelligent medical big data processing method based on artificial intelligence is described in detail below.

And step S110, acquiring medical resource behavior big data of the target subscribed medical resource.

In one embodiment, the target subscribed medical resource may refer to a medical service scenario of practical application, such as a medical drug consultation scenario, a medical department consultation scenario, a medical knowledge point learning scenario, and the like, but is not limited thereto. The medical resource behavior big data may refer to behavior big data generated by a user under the condition that the user subscribes to the medical resource, for example, for a medical knowledge point learning scene, the behavior big data may refer to big data information generated by knowledge point searching behavior, knowledge point sharing behavior and the like of the user.

And step S120, processing the medical resource behavior big data to obtain the search intention information distribution of the medical resource behavior big data.

The search intention information distribution comprises intention distribution information of a target medical pushing label in the medical resource behavior big data, wherein the intention distribution information points to the target in the search. And the search intention information distribution specifically comprises a shared search intention characteristic and a continuously shared search intention characteristic, wherein the shared search intention characteristic comprises a plurality of intention components, and each intention component represents the confidence degree that the behavior interest behavior in the medical resource behavior big data corresponding to the intention component is the intention interest point of the target medical pushing label searched and directed to the target. The dimension of the continuously shared search intention feature is 2, which means that the continuously shared search intention feature is specifically composed of the search intention feature of the feature range attribute and the search intention feature of the feature hierarchy attribute, and the search intention feature of the feature range attribute and the intention keyword interval of the search intention feature of the feature hierarchy attribute are the same. In addition, the intention keyword section sharing the search intention feature is the same as the intention keyword section continuously sharing the search intention feature. For example, both the intention keyword interval sharing the search intention feature and the intention keyword interval continuously sharing the search intention feature are (a 1, a 2.. a.. an.). The search intention characteristics of the characteristic range attribute comprise a plurality of intention components, and each intention component represents a decision characteristic range corresponding to behavior interest behaviors in the medical resource behavior big data corresponding to the intention component; similarly, the search intention feature of the feature level attribute comprises a plurality of intention components, and each intention component represents a decision feature level corresponding to the behavior interest behavior in the medical resource behavior big data corresponding to the intention component.

For example, the representation form of the intent of the embodiment of the present disclosure is a form of sharing + persistent sharing. The search intention information distribution makes decisions on the two, namely sharing the search intention characteristics and continuously sharing the search intention characteristics. The dimension sharing the search intention features is (a 1, a 2.... ann.., an). C, the dimension continuously sharing the intention component features is (a 1, a 2.... ann.. an). 2, and C is the target number of target medical push labels to be decided. Each intention component on the search intention characteristic respectively expresses the confidence of the search pointing to the intention interest point of the target and the decision value of the characteristic range characteristic level at the coordinate position.

Step S130, determining big data mining information of the medical resource behavior big data according to the search intention information distribution, wherein the big data mining information comprises mining label information of a search pointing target of the target medical push label in the medical resource behavior big data.

In one embodiment, mining tag information may include intent objects, which may include, for example: the medical resource behavior big data is decision confidence that the search of the target medical push tag points to the intention interest point of the target, and the medical resource behavior big data is the feature range and the feature level of the intention object corresponding to the search of the target medical push tag.

In one embodiment, the search intent information distribution includes sharing the search intent feature and continuously sharing the search intent feature. The shared search intention features comprise confidence degrees of intention interest points of the target medical pushing label searched by each behavior interest behavior in the medical resource behavior big data, and the continuously shared search intention features comprise feature ranges and feature level data corresponding to each behavior interest behavior in the medical resource behavior big data.

In one embodiment, first, the smart medical system 100 determines the intention interest point of the target for searching the target medical push tag in the medical resource behavior big data according to the shared search intention characteristics. Then, the intelligent medical system 100 determines an intention object of the target in the medical resource behavior big data, which is pointed by the search of the target medical push tag, according to the intention interest point and the feature range and the feature level data corresponding to the behavior interest behavior at the intention interest point. Finally, the smart medical system 100 takes the intention target of the search direction target of the target medical push tag as the mining tag information of the search direction target of the target medical push tag.

For example, each behavioral interest behavior in the medical resource behavior big data corresponds to an intent component in the search intent information distribution. Therefore, big data mining information of the medical resource behavior big data can be determined, and the big data mining information comprises mining label information of a target medical pushing label in the medical resource behavior big data, wherein the target medical pushing label is searched and points to the target.

Subsequently, the smart medical system 100 may label the intention object in the medical resource behavior big data according to the mining tag information of the search direction target of the target medical push tag, for example, the intention interest point of the intention object is (i, j), and since the feature range and the feature level data in the search intention information distribution are converted to the original intention graph keyword interval according to a down-sampling rate (usually 4), the feature range of the intention object in the original medical resource behavior big data is 40, and the feature level is 80. The expression form of the intention object is usually a jump expression form, but the expression form of the intention object may be determined according to a search target of a target medical delivery tag included in the medical resource behavior big data, and if the search target of the target medical delivery tag is a web page, a video stream bullet screen, or the like, the intention object is usually a video stream bullet screen, a web page element, or the like. The representation of the intended object may be flexibly floated according to the search of the target medical push tag, or different business scenarios, which is not specifically limited by the present disclosure. Of course, if the target medical push tag is of more than one type, that is, the target medical push tag includes two or more types, the expression form of each type of the intention object may be different, so as to distinguish different types of search pointing targets in the medical resource behavior big data.

By the medical resource behavior data processing method provided by the embodiment of the disclosure, the medical resource behavior big data of the target subscription medical resource can be processed, so that the search intention information distribution of the medical resource behavior big data is obtained. The search intention information distribution comprises intention distribution information of the target medical pushing label in the medical resource behavior big data, and big data mining information of the medical resource behavior big data is determined according to the search intention information distribution. By the design, the second reference medical resource behavior data of the target subscription medical resource can be trained to obtain the search intention classification unit without marking, and the search intention classification unit can directly process the medical resource behavior data of the target subscription medical resource, so that big data mining information is obtained, marking cost of the medical resource behavior data of the target subscription medical resource can be saved, and efficiency and precision of big data mining are improved.

In one embodiment, the intelligent medical system 100 may invoke the search intention classification unit to process the medical resource behavior big data to obtain the search intention information distribution of the medical resource behavior big data. The search intention classification unit is obtained by performing network weight learning and updating on reference medical resource behavior data of the target subscription medical resource. For example, the search intention classification unit is obtained by training based on first reference medical resource behavior data of the original subscribed medical resource, reference intention distribution information of a search pointing target of a target medical push tag in the first reference medical resource behavior data, and second reference medical resource behavior data of the target subscribed medical resource.

Compared with the prior art, the design is that the second reference medical resource behavior data of the target subscription medical resource can be trained to obtain the search intention classification unit without marking, and the search intention classification unit is obtained by performing network weight learning updating training by using the redundant exclusion information amount of the reference medical resource behavior data of the target subscription medical resource. Finally, the machine learning network obtained through training can directly process the medical resource behavior data of the target subscription medical resource, so that big data mining information is obtained, the marking cost of the medical resource behavior data of the target subscription medical resource can be saved, and the efficiency and the precision of big data mining are improved.

In one embodiment, the search intention classification unit includes a search intention extraction node and a search intention fusion node. For example, a machine learning network includes a search intention extraction node and a search intention fusion node. The specific implementation manner of the intelligent medical system 100 invoking the search intention classification unit to process the medical resource behavior big data to obtain the search intention information distribution of the medical resource behavior big data may include: the intelligent medical system 100 calls the search intention extraction node to perform feature extraction on the medical resource behavior big data so as to obtain initial search intention features of the medical resource behavior big data; and calling the search intention fusion node to perform feature fusion on the medical resource behavior big data and the initial search intention characteristics so as to obtain the search intention information distribution of the medical resource behavior big data. The search intention extraction nodes are composed of convolutional layers, batch regularization, nonlinear activation, pooling layers and the like. The search intention extraction node can effectively extract a feature level dimension feature expression (namely an initial search intention feature) of input medical resource behavior data (medical resource behavior big data).

In one embodiment, first, the smart medical system 100 calls a search intention fusion node to perform convolution processing and sampling processing on medical resource behavior big data and an initial search intention characteristic to obtain a first search intention characteristic; then, the intelligent medical system 100 calls the search intention fusion node to perform compression processing and excitation processing on the medical resource behavior big data and the initial search intention characteristics to obtain influence weights corresponding to the initial search intention characteristics, and performs weighted calculation on the initial search intention characteristics according to the influence weights to obtain second search intention characteristics; finally, the intelligent medical system 100 performs fusion processing on the first to-be-fused search intention feature and the second to-be-fused search intention feature to obtain search intention information distribution of medical resource behavior big data.

For example, the search intention fusion node may include a first fusion structure and a second fusion structure, and the first fusion structure may be, for example, a feature pyramid structure. The feature pyramid structure is a low-level search intention feature representation of an initial search intention feature of the medical resource behavior big data and a feature level layer, and therefore a first search intention feature is obtained. The basic operation unit of the characteristic pyramid structure is also the meta-operation of a convolution layer, batch regularization, nonlinear activation and pooling layer. The second fusion structure may be a compression-excitation module, for example, which may pool the initial search intention features globally and process them by excitation, resulting in impact weights. Finally, the intelligent medical system 100 performs weight comprehensive calculation on the influence weight and the initial search intention feature to obtain a second search intention feature.

Finally, the intelligent medical system 100 performs fusion processing on the first search intention feature and the second search intention feature, so as to obtain search intention information distribution of medical resource behavior big data. Of course, after the initial search intention features of the medical resource behavior big data are processed through the feature pyramid structure, the obtained first search intention features serve as the input of the compression-excitation module, and then the second search intention features are obtained. And finally, taking the second search intention characteristic obtained by processing the compression-excitation module according to the first search intention characteristic as the search intention information distribution of the medical resource behavior big data.

In an independent embodiment, the intelligent medical big data processing method based on artificial intelligence provided by the embodiment of the disclosure comprises the following steps.

Step S210, obtaining a reference data sequence, wherein the reference data sequence comprises first reference medical resource behavior data of an original subscribed medical resource, reference intention distribution information of a target medical push label in the first reference medical resource behavior data, and second reference medical resource behavior data of the target subscribed medical resource.

In one embodiment, the original subscribed medical resources refer to actual business scenarios, and the target subscribed medical resources refer to simulation scenarios. The first reference medical resource behavior data is: medical resource behavior data in an actual business scene.

The first reference medical resource behavior data refers to simulated medical resource behavior data, and may be automatically generated by a professional simulation program, for example. For example, the reference intention distribution information may be an intention object, the expression form of the intention object may be a skip expression form box, and if the target medical push tag is more than one type, the expression forms of different types of intention objects may also be different. It should be noted that the intention object pointing to the target in the search of the target medical pushing tag in the first reference medical resource behavior data is also automatically marked by the simulation software. By the design, the first reference medical resource behavior data and the reference intention distribution information of the target medical pushing label in the first reference medical resource behavior data are automatically synthesized by the computer, and compared with the method of manually acquiring and labeling the medical resource behavior data, the time cost and the labor cost are saved, and the processing efficiency of the medical resource behavior data is improved.

In addition, the second reference medical resource behavior data is medical resource behavior data in an actual application environment. The second medical resource behavior data may be medical resource behavior data arbitrarily selected by the smart medical system 100, and of course, the medical resource behavior data in the medical resource behavior library are all medical resource behavior data in the actual application environment.

Step S220, a reference machine learning unit is trained by performing network weight learning update on the second reference medical resource behavior data and learning on the first reference medical resource behavior data.

Specifically, the network weight learning update refers to adjusting the unit weight of the reference machine learning unit according to the redundancy exclusion information amount of the second reference medical resource behavior data. The learning means that a second target decision evaluation index is calculated according to the first reference medical resource behavior data, and the unit weight of the reference machine learning unit is adjusted according to the second target decision evaluation index.

In one embodiment, the intelligent medical system 100 determines the first objective decision evaluation index of the reference machine learning unit according to the redundancy elimination information amount of the second reference medical resource behavior data. The intelligent medical system 100 determines a second target decision evaluation index of the reference machine learning unit according to the first reference medical resource behavior data and the reference intention distribution information of the search pointing target of the target medical push tag in the first reference medical resource behavior data. The intelligent medical system 100 trains the reference machine learning unit according to the first objective decision evaluation index and the second objective decision evaluation index.

In one embodiment, the intelligent medical system 100 invokes the reference machine learning unit to perform feature extraction on the first reference medical resource behavior data, so as to obtain a first decision intention distribution feature of the first reference medical resource behavior data. Then, the intelligent medical system 100 determines a second target decision evaluation index of the reference machine learning unit according to the first decision intention distribution characteristic and the reference intention distribution information.

In one embodiment, the reference machine learning unit may be, for example, a deep learning network model capable of implementing a target detection function, and aims to find all intention targets with intention tendency in the medical resource behavior data and determine their categories and located service nodes.

In one embodiment, the reference machine learning unit may include a reference search intention extraction node and a reference search intention fusion node. The invoking of the reference machine learning unit by the smart medical system 100 to perform feature extraction on the first reference medical resource behavior data of the originally subscribed medical resource to obtain the first decision intention distribution feature of the first reference medical resource behavior data may include, for example: the intelligent medical system 100 calls the reference search intention extraction node to perform feature extraction on the first reference medical resource behavior data so as to obtain initial search intention features of the first reference medical resource behavior data; the intelligent medical system 100 invokes the reference search intention fusion node to perform feature fusion on the initial search intention features of the first reference medical resource behavior data to obtain first decision intention distribution features of the first reference medical resource behavior data. For example, the structures corresponding to the search intention extraction node and the search intention fusion node respectively can be referred to specifically as the structures corresponding to the search intention extraction node and the search intention fusion node respectively described above. The reference search intention fusion node may further include a first reference structure and a second reference structure, where the structure of the first reference structure may be referred to the first fusion structure specifically, and the structure of the second reference structure may be referred to the second fusion structure specifically.

Of course, the first decision intent distribution feature also includes a first share intent component feature and a first sustained share intent component feature. The first shared intent component feature and the first continuous shared intent component feature are arranged in the same way, for example, both are (a 1, a 2.. a., an.), and of course, the dimension of the first continuous shared intent component feature is 2, which means that the first feature range attribute intent component feature and the first feature level attribute intent component feature are included. In addition, the dimension of the first shared intention component feature is the same as the number of targets included in the target medical push label, for example, if the number of the target medical push labels is 3, the dimension of the first shared intention component feature is also 3, for example, if the number of the target medical push labels is 1, the dimension of the first shared intention component feature is also 1. It should be noted that each intention component in the first decision intention distribution feature includes the same feature meaning as each intention component in the intention interest point search intention feature of the medical resource behavior big data. Namely, the first sharing intention component feature includes a confidence level that each behavior interest behavior in the first reference medical resource behavior data is an intention interest point of the target medical push tag searched for, and the first continuous sharing intention component feature includes a feature range and feature level data corresponding to each behavior interest behavior in the first reference medical resource behavior data.

In one embodiment, the intelligent medical system 100 invokes the reference machine learning unit to perform feature extraction on the second reference medical resource behavior data, so as to obtain a second decision intention distribution feature of the second reference medical resource behavior data. Then, the intelligent medical system 100 determines a first target decision evaluation index of the reference machine learning unit according to the redundant exclusion information amount of the second decision intention distribution feature.

In an embodiment, the intelligent medical system 100 "invokes the reference machine learning unit to perform feature extraction on the second reference medical resource behavior data of the target subscribed medical resource to obtain the second decision intention distribution feature of the second reference medical resource behavior data" in an execution step, specifically, refer to the execution step of the intelligent medical system 100 "invokes the reference machine learning unit to perform feature extraction on the first reference medical resource behavior data of the original subscribed medical resource to obtain the first decision intention distribution feature of the first reference medical resource behavior data" in step S220, which is not described herein again in this disclosure.

It should be noted that, in the training process of the reference machine learning unit based on the first reference medical resource behavior data of the original subscribed medical resource and the second reference medical resource behavior data of the target subscribed medical resource, the first reference medical resource behavior data and the second reference medical resource behavior data are simultaneously input into the reference machine learning unit. In an embodiment, the batch-processed data includes a plurality of first reference medical resource behavior data and an equal number of second reference medical resource behavior data, and of course, the number of the first reference medical resource behavior data and the second reference medical resource behavior data input into the reference machine learning unit may be different in each batch-processing process, which is not specifically limited by the present disclosure.

Step S230, when the trained reference machine learning unit meets the training termination requirement, taking the trained reference machine learning unit as a search intention classification unit, and processing the input medical resource behavior big data based on the search intention classification unit to obtain the search intention information distribution of the medical resource behavior big data.

In one embodiment, the first objective decision evaluation index includes a third decision evaluation index and a fourth decision evaluation index, and the second objective decision evaluation index includes the first decision evaluation index and a second decision evaluation index. The intelligent medical system 100 obtains a first influence weight corresponding to a first decision evaluation index, a second influence weight corresponding to a second decision evaluation index, a third influence weight corresponding to a third decision evaluation index, and a fourth influence weight corresponding to the third decision evaluation index; then, the intelligent medical system 100 performs weight comprehensive calculation on the second target decision evaluation index and the first target decision evaluation index according to the first influence weight, the second influence weight, the third influence weight and the fourth influence weight to obtain a target decision evaluation index; finally, the intelligent medical system 100 adjusts the unit weight of the reference machine learning unit according to the objective decision evaluation index. Subsequently, when the adjusted reference machine learning unit meets the training termination requirement, the adjusted reference machine learning unit is used as a search intention classification unit.

The training termination requirement may be: when the training times of the reference machine learning unit reach a preset training threshold value, for example, 500 times, the reference machine learning unit meets the training termination requirement; when the difference between the decision big data mining information corresponding to each reference medical resource behavior data and the actual big data mining information corresponding to each reference medical resource behavior data is smaller than a difference threshold value, the reference machine learning unit meets the training termination requirement; when the floating value of the decision big data mining information corresponding to each reference medical resource behavior data obtained by the reference machine learning unit through two adjacent training is smaller than the floating threshold value, the reference machine learning unit meets the training termination requirement. The reference medical resource behavior data may be first reference medical resource behavior data or second reference medical resource behavior data.

By the design, the second reference medical resource behavior data of the target subscription medical resource can be trained without marking to obtain the search intention classification unit, and the search intention classification unit can directly process the medical resource behavior data of the target subscription medical resource, so that the big data mining information is obtained. Therefore, the purpose of subscribing medical resource migration can be realized, and the first reference medical resource behavior data is automatically synthesized by simulation and does not need to be collected offline, so that the efficiency of artificial intelligence learning is improved; in addition, the reference intention distribution information of the target medical pushing label in the first reference medical resource behavior data, which points to the target in the search process, is also automatically labeled by the simulation software, so that compared with manual labeling, the efficiency and the accuracy are improved, the accuracy of artificial intelligence learning is further improved, and the efficiency of the artificial intelligence learning is further improved.

In a separate embodiment, the disclosed embodiment provides a flow of a method for determining a first objective decision evaluation index, which is applied to the intelligent medical system 100, and the method corresponds to an embodiment of step S220, which includes the following steps.

Step S310, performing feature vector unification on the second sharing intention component feature or the second continuous sharing intention component feature included in the second decision intention distribution feature to obtain a second sharing intention component feature or a second continuous sharing intention component feature after the vector unification.

In one embodiment, the intelligent medical system 100 processes the second reference medical resource behavior data by invoking the reference machine learning unit to obtain a second decision intention distribution characteristic of the second reference medical resource behavior data. Wherein the second decision intent distribution feature comprises a second sharing intent component feature and a second persistent sharing intent component feature. The intelligent medical system 100 may unify the feature vectors of the second shared intention component features to obtain the second shared intention component features after the feature vectors are unified. Similarly, the intelligent medical system 100 may also unify the feature vectors of the second continuous shared intention component features to obtain the second continuous shared intention component features after the feature vectors are unified.

Step S320, calculating a redundant exclusion information amount for each of the plurality of intention components, and obtaining a third decision evaluation index according to the redundant exclusion information amounts of all the intention components, the feature range of the second decision intention distribution feature, and the feature level.

In one embodiment, the second shared intent component feature after vector unification includes a plurality of intent components, each intent component corresponding to a behavior interest behavior in the second reference medical resource behavior data. The intelligent medical system 100 calculates a redundancy exclusion information amount (information entropy) for each of the plurality of intention components.

Step S330, calculating a maximum square decision evaluation index for each of the plurality of intent components, and obtaining a fourth decision evaluation index according to the maximum square decision evaluation indexes of all the intent components, the feature range of the second decision intent distribution feature, and the feature level.

In one embodiment, the second shared intent component feature after vector unification includes a plurality of intent components, each intent component corresponding to a behavior interest behavior in the second reference medical resource behavior data. The intelligent medical system 100 calculates a maximum square decision evaluation index for each of the plurality of intention components.

Step S340, determining a first target decision evaluation index of the reference machine learning unit according to the third decision evaluation index and the fourth decision evaluation index.

In one embodiment, the intelligent medical system 100 obtains a third influence weight corresponding to a third decision evaluation index and obtains a fourth influence weight corresponding to a fourth decision evaluation index. Then, the intelligent medical system 100 performs weight comprehensive calculation on the third decision evaluation index and the fourth decision evaluation index according to the third influence weight and the fourth influence weight to obtain a first target decision evaluation index of the reference machine learning unit.

Finally, the intelligent medical system 100 performs a superposition operation on the second objective decision evaluation index and the first objective decision evaluation index to obtain an objective decision evaluation index.

In a separate embodiment, the present disclosure provides a method for determining a second objective decision evaluation indicator, which is applied to the intelligent medical system 100, and corresponds to an embodiment of step S220, and the method includes the following steps.

Step S410, determining a first decision evaluation index according to the first sharing intention component characteristic, the intention interest point of the target medical pushing label, and the quantity of the first reference medical resource behavior data.

The reference intention distribution information of the target medical pushing tag in the first reference medical resource behavior data specifically includes a characteristic range and a characteristic hierarchy of an intention object of the target medical pushing tag in the first reference medical resource behavior data, and an intention interest point of the target medical pushing tag in the first reference medical resource behavior data.

In one embodiment, the intelligent medical system 100 processes the first reference medical resource behavior data by invoking the reference machine learning unit to obtain a first decision intention distribution characteristic of the first reference medical resource behavior data. Wherein the first decision intent distribution feature comprises a first shared intent component feature comprising a confidence level that each behavioral interest behavior in the first reference medical resource behavior data is an intent point of interest to a target for a search of a target medical push tag.

Step S420, determining a second decision evaluation index according to the first continuous sharing intention component characteristic, the quantity of the first reference medical resource behavior data, the characteristic range and the characteristic hierarchy of the intention object.

In one embodiment, the intelligent medical system 100 processes the first reference medical resource behavior data by invoking the reference machine learning unit to obtain a first decision intention distribution characteristic of the first reference medical resource behavior data. Wherein the first decision intent distribution feature comprises a first continuous sharing intent component feature comprising a feature range and feature level data corresponding to each behavioral interest behavior in the first reference medical resource behavior data. The reference intention distribution information of the target medical pushing tag in the first reference medical resource behavior data includes a characteristic range, a characteristic hierarchy and an intention interest point of the target medical pushing tag in the first reference medical resource behavior data.

Step S430, determining a second target decision evaluation index of the reference machine learning unit according to the first decision evaluation index and the second decision evaluation index.

In one embodiment, the intelligent medical system 100 obtains a first impact weight corresponding to a first decision evaluation index and obtains a second impact weight corresponding to a second decision evaluation index. Then, the intelligent medical system 100 performs weight comprehensive calculation on the first decision evaluation index and the second decision evaluation index according to the first influence weight and the second influence weight to obtain a second target decision evaluation index of the reference machine learning unit.

In one embodiment, the embodiment of the present disclosure further provides a medical information pushing method based on smart medical big data, which includes the following steps.

And A110, acquiring E-commerce hotspot information of each relevant mining tag in a corresponding target medical E-commerce data source according to the big data mining information.

In this embodiment, the e-commerce hotspot information of each relevant mining tag can be obtained through the information acquisition service of each relevant mining tag. The e-commerce hotspot information can comprise e-commerce information hotspot information, recent scientific research hotspot information and the like.

And A120, acquiring the E-commerce hotspot distribution characteristics of the relevant mining tags according to the E-commerce hotspot information of the relevant mining tags.

In this embodiment, feature extraction may be performed through a pre-trained AI model, and the extracted features are subjected to processing such as screening and fusion to obtain the e-commerce hotspot distribution feature.

Step A130, fusing the E-commerce hotspot distribution characteristics of the relevant mining labels to obtain mining hotspot distribution characteristics of the target medical E-commerce data source.

In this embodiment, the mining hotspot distribution characteristics of the whole target medical e-commerce data source are obtained by fusing the e-commerce hotspot distribution characteristics of the relevant mining tags, so that the mining hotspot distribution characteristics can reflect the influence of e-commerce hotspot information of the relevant mining tags in the medical e-commerce data source on the whole medical e-commerce data source, and the decision characteristics of the push attribute of the target medical e-commerce data source can be more accurately expressed.

Step A140, performing medical e-commerce push attribute decision on the target medical e-commerce data source according to the mining hotspot distribution characteristics to obtain at least one medical e-commerce push attribute of the target medical e-commerce data source.

Step a150, performing corresponding medical e-commerce information pushing on the intelligent medical subscription device 200 according to at least one medical e-commerce pushing attribute of the target medical e-commerce data source.

In one embodiment, the medical e-commerce push attribute decision of the target medical e-commerce data source may be implemented in step a140 by a classifier, such as a support vector machine or a fully connected deep neural network.

In this embodiment, since the mining hotspot distribution characteristics are obtained by fusing the e-commerce hotspot distribution characteristics of the relevant mining tags, the medical e-commerce push attribute decision of the target medical e-commerce data source executed according to the mining hotspot distribution characteristics can be more accurate.

In one embodiment, step a120 may include the following substeps.

Step A210, for the e-commerce hotspot information of each relevant mining tag, performing hotspot public opinion information extraction on the e-commerce hotspot information of the relevant mining tag to obtain hotspot public opinion information corresponding to the e-commerce hotspot information of the relevant mining tag.

In this embodiment, the e-commerce hotspot information of each relevant mining tag is respectively input into a pre-trained AI model, and one or more convolution processes are performed through the AI model to perform feature extraction on the e-commerce hotspot information, so as to obtain hotspot public opinion information corresponding to the e-commerce hotspot information of the relevant mining tag.

Step A220, performing public sentiment trend characteristic identification on the hot public sentiment information of the e-commerce hotspot information of the relevant mining label, and determining at least one public sentiment trend characteristic of the hot public sentiment information of the e-commerce hotspot information of the relevant mining label.

The E-commerce hotspot information of a single relevant mining tag may have some invalid features, such as universality features; or have some behavior that is too personalized to characterize the hot spot push attributes of the medical e-commerce data source. Therefore, in the embodiment, representative public opinion trend features need to be identified from the hot public opinion information of a single relevant mining label, and the features corresponding to the public opinion trend features will influence the subsequent feature extraction and fusion process.

Step A230, performing hotspot extraction on each public opinion trend feature in the hotspot public opinion information of the e-commerce hotspot information of the relevant mining label to obtain the public opinion trend hotspot features of each public opinion trend feature of the e-commerce hotspot information of the relevant mining label.

In this embodiment, after the public opinion trend features are identified, specific feature vectors corresponding to the public opinion trend features in the hotspot public opinion information may be extracted and hotspot extraction may be performed to obtain the public opinion trend hotspot features corresponding to the public opinion trend features.

Step A240, according to the preset weighting information of the decision result of the target medical e-commerce data source of the e-commerce hotspot information of the relevant mining label, fusing the hotspot public opinion information of the e-commerce hotspot information of the relevant mining label and the public opinion trend hotspot characteristics of the public opinion trend characteristics to obtain the e-commerce hotspot distribution characteristics of the relevant mining label.

In this embodiment, after the public opinion trend hotspot characteristic is determined, the public opinion trend hotspot characteristic and the hotspot public opinion information may be weighted and fused according to the importance of the public opinion trend hotspot characteristic to the medical e-commerce push attribute decision of the medical e-commerce data source. Therefore, the obtained E-commerce hotspot distribution characteristics comprise the global distribution characteristics and the local distribution characteristics of the related mining tags, and the hotspot distribution characteristics of the related mining tags can be more accurately reflected.

In one embodiment, step a240 may include the following substeps.

Step A241, determining fusion parameters corresponding to all public sentiment trend characteristics of the E-commerce hotspot information of the relevant mining labels according to preset weighted information of all public sentiment trend characteristics of the E-commerce hotspot information of the relevant mining labels on the decision result of the target medical E-commerce data source.

In this embodiment, different public opinion trend characteristics may have different preset weighting information, and the fusion parameter may be determined by an influence degree of the public opinion trend characteristics on medical e-commerce push attribute decision for target medical e-commerce data source classification.

Step A242, according to the fusion parameters, performing weighted calculation on the hot spot public sentiment information of the e-commerce hotspot information of the related mining labels and the public sentiment trend hotspot characteristics of all the public sentiment trend characteristics to obtain e-commerce hotspot distribution characteristics of the related mining labels.

In this embodiment, the public opinion information of the e-commerce hotspot information of the relevant mining tag and the public opinion trend hotspot features of each public opinion trend feature are weighted and calculated, so that the obtained e-commerce hotspot distribution features have both the hotspot public opinion information representing the global features of the relevant mining tag and the public opinion trend hotspot features having the characteristic hotspot distribution features, and the public opinion trend hotspot features can reflect the importance of some public opinion trend features through weighting calculation. Therefore, the obtained E-commerce hotspot distribution characteristics can more accurately reflect hotspot distribution characteristics of relevant mining tags, which can influence the target medical E-commerce data source.

In one embodiment, step a130 may include the following sub-steps.

Step A131, clustering the e-commerce hotspot distribution characteristics of the relevant excavation tags to obtain at least one cluster, and determining the core distribution characteristics serving as a clustering core in each cluster.

Step A132, for each cluster, calculating hot spot distribution node characteristics of the non-core distribution characteristics and the core distribution characteristics in the cluster to obtain a hot spot distribution node characteristic sequence of the cluster.

Step A133, fusing the characteristic sequences of the hot spot distribution nodes of each cluster to obtain the mining hot spot distribution characteristics of the target medical e-commerce data source.

In this embodiment, the clustering calculation may be performed by using a K-means (K-means) clustering algorithm.

In one embodiment, step a131 may include the following substeps.

Step 1311, determining the number N of clusters, where N is a positive integer greater than or equal to 2.

Step 1312, selecting N e-commerce hot spot distribution characteristics from the e-commerce hot spot distribution characteristics of the relevant mining tags to serve as N clustered core distribution characteristics respectively.

And 1313, calculating feature similarity between the e-commerce hotspot distribution features of the relevant mining tags and the core distribution features.

In this embodiment, the feature similarity between the e-commerce hotspot distribution feature and the core distribution feature may represent a matching degree between the e-commerce hotspot distribution feature and the core distribution feature. The greater the feature similarity, the greater the degree of matching. The way of calculating the feature similarity between the e-commerce hotspot distribution feature and the core distribution feature may be calculated by a cosine distance or a euclidean distance, etc.

Step 1314, adding each e-commerce hotspot distribution characteristic to the cluster to which the core distribution characteristic with the largest characteristic similarity with the e-commerce hotspot distribution characteristic belongs to respectively to obtain N clusters.

And 1315, selecting the e-commerce hotspot distribution characteristics meeting the clustering core conditions from the clusters as new core distribution characteristics for each cluster, and returning to the step of calculating the feature similarity between the e-commerce hotspot distribution characteristics of each relevant mining tag and each core distribution characteristic until the core distribution characteristics of each cluster meet the clustering termination requirement, so as to obtain N clusters, and obtain the core distribution characteristics serving as clustering cores in each cluster.

In this embodiment, for each cluster, whether the latest core distribution feature of the cluster is the same as the core distribution feature used at the maximum time in the clustering process, that is, whether the feature similarity between the two is 0, is calculated. If the cluster cores are the same, the cluster cores of the cluster are not changed, if all the cluster cores of the cluster are not changed, the clustering process is completed, N clusters are obtained, and the core distribution characteristics serving as the cluster cores in each cluster are obtained; if not all clustered cores have changed, then return to step A1313 until each clustered core has no more changes.

It should be understood that the latest core distribution characteristic of each cluster in the cluster calculation is the same as the cluster core used by the cluster at the maximum time, which is only an optional condition for ending the loop, and the optional condition may also be that the hot spot distribution node characteristic between the two cluster cores is smaller than a certain preset value.

In one embodiment, in step a220, when hot spot public opinion information is extracted from the e-commerce hotspot information of the relevant mining tag to obtain hot spot public opinion information corresponding to the e-commerce hotspot information of the relevant mining tag, the hot spot public opinion information may be extracted from the e-commerce hotspot information of the relevant mining tag through a deep learning network model to obtain hot spot public opinion information corresponding to the e-commerce hotspot information of the relevant mining tag.

In step a220, performing public opinion trend feature recognition on the hot spot public opinion information of the e-commerce hotspot information of the relevant mining tag, and when determining at least one public opinion trend feature of the hot spot public opinion information of the e-commerce hotspot information of the relevant mining tag, performing public opinion trend feature recognition on the hot spot public opinion information of the e-commerce hotspot information of the relevant mining tag through the deep learning network model, and determining at least one public opinion trend feature of the hot spot public opinion information of the e-commerce hotspot information of the relevant mining tag.

In step a140, when a medical e-commerce push attribute decision is made on the target medical e-commerce data source according to the mining hotspot distribution characteristics to obtain at least one medical e-commerce push attribute of the target medical e-commerce data source, the medical e-commerce push attribute decision can be made on the target medical e-commerce data source according to the mining hotspot distribution characteristics through the deep learning network model to obtain at least one medical e-commerce push attribute of the target medical e-commerce data source.

In this embodiment, the deep learning network model may be a residual network, a densely connected convolutional network, or the like.

In one embodiment, the embodiment of the present disclosure further provides a medical information pushing method based on smart medical big data, which includes the following steps.

Step A401, a reference data set is obtained, wherein the reference data set comprises e-commerce hotspot information of a target medical e-commerce data source and actual medical e-commerce push attributes corresponding to the target medical e-commerce data source.

Step A402, performing hotspot public opinion information extraction on the E-commerce hotspot information of the target medical E-commerce data source through a deep learning network model to obtain hotspot public opinion information corresponding to the E-commerce hotspot information of the target medical E-commerce data source, performing public opinion trend feature recognition on the hotspot public opinion information of the E-commerce hotspot information of the target medical E-commerce data source, and determining at least one decision public opinion trend feature of the hotspot public opinion information of the E-commerce hotspot information of the target medical E-commerce data source.

Step a403, performing feature extraction on each decision public opinion trend feature in the hot point public opinion information of the e-commerce hotspot information of the target medical e-commerce data source to obtain a public opinion trend hotspot feature of each decision public opinion trend feature of the e-commerce hotspot information of the target medical e-commerce data source, and fusing the hot point public opinion information of the e-commerce hotspot information of the target medical e-commerce data source and the public opinion trend hotspot feature of each decision public opinion trend feature according to each decision public opinion trend feature of the e-commerce hotspot information of the target medical e-commerce data source to obtain an e-commerce hotspot distribution feature of the e-commerce hotspot information of the target medical e-commerce data source.

Step A404, fusing E-commerce hotspot distribution characteristics of E-commerce hotspot information of each target medical E-commerce data source to obtain mining hotspot distribution characteristics of the target medical E-commerce data source.

Step A405, determining the matching probability of the target medical e-commerce data source on each preset medical e-commerce pushing attribute according to the mining hotspot distribution characteristics.

Step A406, calculating a first model evaluation index value between the matching probability and the actual medical e-commerce pushing attribute of the target medical e-commerce data source.

Step A407, calculating a gradient descending value of the first model evaluation index value to the mining hotspot distribution characteristic of the target medical e-commerce data source, and calculating public opinion probability distribution corresponding to hotspot public opinion information of the e-commerce hotspot information of the target medical e-commerce data source according to the gradient descending value.

Step A408, determining medical e-commerce push attribute information of the target medical e-commerce data source according to the matching probability of the target medical e-commerce data source.

Step A409, when the medical e-commerce pushing attribute information of the target medical e-commerce data source is consistent with the actual medical e-commerce pushing attribute, acquiring public opinion trend characteristics of hotspot public opinion information of e-commerce hotspot information of the target medical e-commerce data source according to the public opinion probability distribution, and setting the acquired public opinion trend characteristics as the actual public opinion trend characteristics of the e-commerce hotspot information of the target medical e-commerce data source.

Step A410, when the medical e-commerce pushing attribute information of the target medical e-commerce data source is not matched with the actual medical e-commerce pushing attribute, acquiring non-public opinion trend characteristics of hot public opinion information of e-commerce hot spot information of the target medical e-commerce data source according to the public opinion probability distribution, and setting the acquired non-public opinion trend characteristics as the non-actual public opinion trend characteristics of the e-commerce hot spot information of the target medical e-commerce data source.

Step A411, calculating a second model evaluation index value of the decision public opinion trend characteristic of the E-commerce hotspot information of the target medical E-commerce data source according to the actual public opinion trend characteristic and the non-actual public opinion trend characteristic.

Step A412, adjusting parameters of the deep learning network model according to the first model evaluation index value and the second model evaluation index value to obtain the deep learning network model meeting the conditions.

In this embodiment, a back propagation algorithm may be adopted to adjust parameters of the deep learning network model, so that a first model evaluation index value between the matching probability obtained through the deep learning network model and the actual medical e-commerce push attribute is smaller than a target feature degree, where the target feature degree may be set as small as possible, so as to improve the classification accuracy of the deep learning network model.

Generally, if the matching probability of the deep learning network model on a certain preset medical e-commerce pushing attribute exceeds a threshold value, the target medical e-commerce data source can be considered as the medical e-commerce data source on the preset medical e-commerce pushing attribute. In the training process of the deep learning network model, if the pushing attribute information of the medical e-commerce decided by the deep learning network model is consistent with the pushing attribute of the actual medical e-commerce, namely the decision is correct, public opinion probability distribution can be obtained through analysis according to the parameters related in the decision process, public opinion trend feature recognition can be carried out according to the public opinion probability distribution, and the actual public opinion trend feature of the e-commerce hotspot information of the target medical e-commerce data source is obtained.

In the training process of the deep learning network model, if the pushing attribute information of the medical e-commerce decided by the deep learning network model is not matched with the pushing attribute of the actual medical e-commerce, namely, the pushing attribute of the medical e-commerce of the target medical e-commerce data source is decided wrongly by the deep learning network model, public opinion probability distribution can be obtained through analysis according to the parameters related in the decision process, and non-actual public opinion trend characteristics of the e-commerce hotspot information of the target medical e-commerce data source can be obtained according to the public opinion probability distribution.

Specifically, in the present embodiment, step a411 may include the following sub-steps.

Step A4112, determining the probability of the non-actual public opinion trend characteristic of the decision public opinion trend characteristic according to the decision public opinion trend characteristic of the E-commerce hotspot information of the target medical E-commerce data source and the behavior similarity of the actual public opinion trend characteristic, and according to the decision public opinion trend characteristic of the E-commerce hotspot information of the target medical E-commerce data source and the behavior similarity of the non-actual public opinion trend characteristic.

Optionally, in an embodiment, the probability of the actual public opinion trend feature of the decision public opinion trend feature having a feature matching degree with the actual public opinion trend feature larger than the first target feature may be set to 1; setting the probability of the actual public opinion tendency feature of the decision public opinion tendency feature with the feature matching degree of the non-actual public opinion tendency feature larger than the second target feature degree to 0; the first target feature degree and the second target feature degree can be set according to actual conditions.

Step A4113, determining the matching probability of the decision public opinion trend characteristic as the actual public opinion trend characteristic according to the public opinion trend hotspot characteristic of the decision public opinion trend characteristic through a deep learning network model.

Step A4114, calculating a decision difference parameter of the decision public opinion trend characteristic according to the matching probability of the decision public opinion trend characteristic and the corresponding actual public opinion trend characteristic probability.

Step A4115, calculating regression difference parameters of the decision public opinion trend characteristics according to the decision public opinion trend characteristics that the actual public opinion trend characteristic probability is not lower than a preset probability threshold, the service nodes in the hot public opinion information of the E-commerce hotspot information of the target medical E-commerce data source and the service nodes in the hot public opinion information of the E-commerce hotspot information of the target medical E-commerce data source of the actual public opinion trend characteristics.

Step A4116, the decision difference parameter and the regression difference parameter are fused to obtain a second model evaluation index value of the decision public opinion trend characteristic of the E-commerce hotspot information of the target medical E-commerce data source.

In one embodiment, public opinion probability distribution can be obtained through analysis of a class activation graph weighted by a gradient descending value, a gradient descending value of a first model evaluation index value to a feature of mining hotspot distribution of a target medical e-commerce data source is calculated, a weight parameter corresponding to each hotspot public opinion object in hotspot public opinion information of e-commerce hotspot information of the target medical e-commerce data source is calculated by using a global average of the gradient descending values, and public opinion probability distribution corresponding to the hotspot public opinion information can be drawn according to the weight parameter of each hotspot public opinion object in the hotspot public opinion information. The mining hotspot distribution characteristics of the target medical e-commerce data source can be obtained by splicing e-commerce hotspot distribution characteristics corresponding to e-commerce hotspot information of each relevant mining tag of the target medical e-commerce data source. The basic idea of the Grad-CAM is that the weight parameter of the feature map corresponding to a certain decision attribute can be converted to express this weight parameter by a back-propagation gradient descent value.

Therefore, by extracting the E-commerce hotspot information of each relevant mining tag in the target medical E-commerce data source, identifying public opinion trend characteristics from the E-commerce hotspot information of a single relevant mining tag, determining the E-commerce hotspot distribution characteristics of each relevant mining tag according to the public opinion trend characteristics, then fusing the E-commerce hotspot distribution characteristics of each relevant mining tag to obtain the mining hotspot distribution characteristics of the whole target medical E-commerce data source, and then performing medical E-commerce push attribute decision on the target medical E-commerce data source according to the mining hotspot distribution characteristics. The determined mining hotspot distribution characteristics can more accurately reflect the influence of the related mining tags on the medical e-commerce data source, so that the characteristics of the whole medical e-commerce data source are more accurately reflected, and the medical e-commerce push attribute decision of the medical e-commerce data source based on the mining hotspot distribution characteristics is more accurate.

Fig. 3 is a schematic diagram of functional modules of an artificial intelligence based intelligent medical big data processing apparatus 300 according to an embodiment of the disclosure, and the artificial intelligence based intelligent medical big data processing apparatus 300 can be a computer program (including program code) running in the intelligent medical system 10010, for example, the artificial intelligence based intelligent medical big data processing apparatus 300 is an application software, and the functions of the functional modules of the artificial intelligence based intelligent medical big data processing apparatus 300 are described in detail below.

The obtaining module 310 is configured to subscribe to a medical resource for a target of any one of the intelligent medical subscription devices 200, and obtain medical resource behavior big data of the target subscribed medical resource.

The processing module 320 is configured to process the medical resource behavior big data to obtain search intention information distribution of the medical resource behavior big data, where the search intention information distribution includes intention distribution information of a search pointing target of a target medical push tag in the medical resource behavior big data.

The determining module 330 is configured to determine big data mining information of the medical resource behavior big data according to the search intention information distribution, where the big data mining information includes mining tag information that a search of a target medical push tag in the medical resource behavior big data points to a target.

Fig. 4 is a schematic diagram illustrating a hardware structure of an intelligent medical system 100 for implementing the artificial intelligence based intelligent medical big data processing method according to the embodiment of the disclosure, and as shown in fig. 4, the intelligent medical system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 may execute the intelligent medical big data processing method based on artificial intelligence according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 may be configured to control transceiving actions of the communication unit 140, so as to perform data transceiving with the intelligent medical subscription device 200.

The specific implementation process of the processor 110 can be seen in the above-mentioned embodiments of the method performed by the intelligent medical system 100, which have similar implementation principles and technical effects, and the detailed description of the embodiments is omitted here.

In addition, the embodiment of the disclosure also provides a readable storage medium, wherein a computer executing instruction is preset in the readable storage medium, and when a processor executes the computer executing instruction, the above intelligent medical big data processing method based on artificial intelligence is realized.

While the disclosure has been described with reference to specific exemplary embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the disclosure.