1. A conversation method comprising:

executing a specific task corresponding to a current user response based on the current user response input by a user;

if the execution result of the specific task triggers dialogue recommendation, determining a current recommendation type corresponding to the current user response in multiple recommendation types;

and generating a current recommendation response corresponding to the current recommendation type based on the current recommendation type, and feeding back the current recommendation response to the user.

2. The method of claim 1, wherein the method further comprises:

if the execution result is that the specific task is successful and the specific task has a continuing task, determining that the execution result of the specific task triggers dialogue recommendation; alternatively, the first and second electrodes may be,

and if the execution result is that the specific task fails, determining that the execution result of the specific task triggers dialogue recommendation.

3. The method of claim 1, wherein the determining a current recommendation type corresponding to the current user response comprises:

and classifying the dialog information including the current user response by adopting a pre-trained classification model so as to determine the current recommendation type corresponding to the current user response.

4. The method according to any one of claims 1-3, wherein the generating a recommendation response corresponding to the current recommendation type based on the current recommendation type comprises:

and generating a current recommendation response corresponding to the current recommendation type based on the recommendation template corresponding to the current recommendation type and/or including the dialog information responded by the current user.

5. The method according to claim 4, wherein the generating a current recommendation response corresponding to the current recommendation type based on the recommendation template corresponding to the current recommendation type and/or including dialog information of the current user response comprises:

if the current recommendation type is a new guide task type, acquiring a recommendation template corresponding to the new guide task type, processing the recommendation template corresponding to the new guide task type by adopting the dialogue information to obtain a processed recommendation template, and taking the processed recommendation template as the current recommendation response; alternatively, the first and second electrodes may be,

and if the current recommendation type is the current task completion guidance type, acquiring a recommendation template corresponding to the current task completion guidance type, and taking the recommendation template corresponding to the current task completion guidance type as the current recommendation response.

6. A conversation apparatus comprising:

the execution module is used for executing a specific task corresponding to the current user response based on the current user response input by the user;

the determining module is used for determining a current recommendation type corresponding to the current user response in multiple recommendation types if the execution result of the specific task triggers the dialogue recommendation;

and the recommending module is used for generating a current recommending response corresponding to the current recommending type based on the current recommending type and feeding the current recommending response back to the user.

7. The apparatus of claim 6, wherein the apparatus further comprises:

the triggering module is used for determining that the execution result of the specific task triggers the dialogue recommendation if the execution result is that the specific task is successful and the specific task has a continuing task; or if the execution result is that the specific task fails, determining that the execution result of the specific task triggers dialog recommendation.

8. The apparatus of claim 6, wherein the determining module is specifically configured to:

and classifying the dialog information including the current user response by adopting a pre-trained classification model so as to determine the current recommendation type corresponding to the current user response.

9. The apparatus according to any one of claims 6-9, wherein the recommendation module is specifically configured to:

and generating a current recommendation response corresponding to the current recommendation type based on the recommendation template corresponding to the current recommendation type and/or including the dialog information responded by the current user.

10. The apparatus of claim 9, wherein the recommendation module is further specifically configured to:

if the current recommendation type is a new guide task type, acquiring a recommendation template corresponding to the new guide task type, processing the recommendation template corresponding to the new guide task type by adopting the dialogue information to obtain a processed recommendation template, and taking the processed recommendation template as the current recommendation response; alternatively, the first and second electrodes may be,

and if the current recommendation type is the current task completion guidance type, acquiring a recommendation template corresponding to the current task completion guidance type, and taking the recommendation template corresponding to the current task completion guidance type as the current recommendation response.

11. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

12. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-5.

13. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-5.

Background

Human-Machine Conversation (Human-Machine Conversation) refers to a technique for a Machine to understand and use natural language to realize Human-Machine communication. A task-based dialog system is a man-machine dialog system aimed at completing tasks. With the development of technology, a man-machine conversation system can meet task requirements of users more and more, but a task failure condition also exists.

In the related art, when a task fails, information of fixed content is generally fed back to a user.

Disclosure of Invention

The present disclosure provides a dialogue method, apparatus, device and storage medium.

According to an aspect of the present disclosure, there is provided a dialogue method including: executing a specific task corresponding to a current user response based on the current user response input by a user; if the execution result of the specific task triggers dialogue recommendation, determining a current recommendation type corresponding to the current user response in multiple recommendation types; and generating a current recommendation response corresponding to the current recommendation type based on the current recommendation type, and feeding back the current recommendation response to the user.

According to another aspect of the present disclosure, there is provided a conversation apparatus including: the execution module is used for executing a specific task corresponding to the current user response based on the current user response input by the user; the determining module is used for determining a current recommendation type corresponding to the current user response in multiple recommendation types if the execution result of the specific task triggers the dialogue recommendation; and the recommending module is used for generating a current recommending response corresponding to the current recommending type based on the current recommending type and feeding the current recommending response back to the user.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of the above aspects.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method according to any one of the above aspects.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of the above aspects.

According to the technical scheme of the disclosure, more flexible recommendation response can be generated.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram according to a first embodiment of the present disclosure;

FIG. 2 is a schematic diagram according to a second embodiment of the present disclosure;

FIG. 3 is a schematic diagram according to a third embodiment of the present disclosure;

FIG. 4 is a schematic diagram according to a fourth embodiment of the present disclosure;

FIG. 5 is a schematic diagram according to a fifth embodiment of the present disclosure;

FIG. 6 is a schematic diagram according to a sixth embodiment of the present disclosure;

FIG. 7 is a schematic diagram according to a seventh embodiment of the present disclosure;

FIG. 8 is a schematic diagram according to an eighth embodiment of the present disclosure;

FIG. 9 is a schematic diagram according to a ninth embodiment of the present disclosure;

FIG. 10 is a schematic diagram according to a tenth embodiment of the present disclosure;

FIG. 11 is a schematic diagram according to an eleventh embodiment of the present disclosure;

fig. 12 is a schematic diagram of an electronic device for implementing any one of the conversation methods of the disclosed embodiments.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The inventor finds out the following reasons mainly for task-based dialog failure through the analysis of task-based dialog:

(1) the generalization capability of the dialogue understanding model is insufficient, so that some user responses which are obviously within the capability range of the system cannot be correctly recognized by the system, and the trend fails.

(2) The content of the user query exceeds the capability boundary of the current system, resulting in a session failure.

(3) The user can use the human-to-human communication mode in daily life in the man-machine conversation, and some non-task conversation contents are introduced, such as: the spoken expressions of "haha", "i think of", "you guess", etc. cause the system to fail to understand correctly.

(4) The user is completely unaware of the capability boundaries of the current dialog system and dialogs are conducted with a trial-and-error attitude, which is particularly prone to dialog failures without explicit intent.

In the related art, when a task fails, a human-computer conversation system (which may be simply referred to as a system) may feed back information of fixed content to a user, or may be referred to as a feedback fixed language, for example, "i cannot understand your meaning" generally. However, such a single fixed telephone technique cannot be well adapted to the specific reason of task failure, and cannot well guide the subsequent operations of the user, which affects the user experience.

In order to provide richer and flexible recommendation responses, applicable to more scenarios, the present disclosure may provide the following embodiments.

Fig. 1 is a schematic diagram according to a first embodiment of the present disclosure. The embodiment provides a dialogue method, which comprises the following steps:

101. and executing the specific task corresponding to the current user response based on the current user response input by the user.

102. And if the execution result of the specific task triggers dialogue recommendation, determining a current recommendation type corresponding to the current user response in multiple recommendation types.

103. And generating a current recommendation response corresponding to the current recommendation type based on the current recommendation type, and feeding back the current recommendation response to the user.

The execution subject of the embodiment may be referred to as a man-machine conversation system, and the man-machine conversation system may be located on a server or a terminal device. Further, the human-machine dialog system may be a task-based human-machine dialog system, targeting the completion of a specific task.

Taking the example that the man-machine conversation system is located on the server, a client for man-machine conversation may be installed on a terminal device used by the user, where the client is, for example, in the form of an Application (APP). The user can input information into the client in the form of voice, text and the like, the information input by the user can be called a user response, then the client can send the user response to a man-machine conversation system located on the server, the man-machine conversation system can feed back to the user based on the user response input by the user, and the information fed back by the man-machine conversation system can be called a system response.

Further, the system response may be a task response or a recommendation response, where the task response refers to a normal response fed back by the system when the system completes a specific task corresponding to the user response, and the recommendation response refers to content recommended to the user based on the current scene.

One user response and one system response can form a round of conversation process, and a task type conversation process generally comprises a plurality of rounds of conversation processes. In this embodiment, a task response of the current round or a recommended response of the current round may be obtained based on a user response of the current round corresponding to each round of the conversation process, and the task response or the recommended response may be fed back to the user as a system response. The user response of the current round can be called as a current user response, and the task response and the recommended response of the current round can be called as a current task response and a current recommended response respectively.

The specific task refers to a specific task corresponding to the current user response, and the task includes, for example: ordering drinks, ordering air tickets, inquiring weather and the like. For example, if the current user response is about ordering a drink, the specific task is ordering a drink.

After receiving the user response input by the user, the system may perform a corresponding specific task based on the user response, for example, perform a task of ordering drinks based on the content about ordering drinks input by the user.

The execution results of the specific task include: the task is successful, or the task is failed, the task success means that the user is finished responding to the corresponding requirement, and the task failure means that the user is not finished responding to the corresponding requirement. Generally speaking, when a task is executed, a task intention and a corresponding word slot are identified, if the task intention is successfully identified and filling of the corresponding word slot is completed, the task is successful, otherwise, the task is failed. Taking beverage ordering as an example, supposing that the word slot corresponding to the task includes the type and the number of the beverage, supposing that the user response output by the user includes that 'I wants to drink milk tea', the task intention can be recognized as beverage ordering, the type of the beverage is milk tea, at this time, because the number word slot is also needed, the system can feed back 'needs several cups' to the user, the user response input again by the user can be '1 cup', after the system recognizes that the number word slot is 1, the filling of the needed word slot is completed, at this time, the word slot needing to be filled is completely filled, the success of the task can be determined, and then the system can feed back the system response corresponding to the success of the task to the user, such as 'good, order placing success'. In the above interaction process, the "need of several cups" fed back to the user by the system is the content fed back for completing the task, and the system response is the task response.

After the system gives feedback to the user that "several cups are needed", if the user does not make a response similar to "1 cup" above, but says something else, for example, the user responds that "today's weather is good" and the system cannot complete word slot filling based on the user response, it can be determined that the task failed. It is to be understood that the above determination process of task success or task failure is exemplary, and the actual man-machine conversation process is more complex, and the task success or task failure can be determined according to the actual scene.

In the related art, after the task fails, the system feeds back a pre-configured fixed dialog to the user, for example, "i can not understand your meaning".

In contrast, in the embodiment of the present disclosure, there may be multiple types of recommendation responses, and a recommendation response suitable for a current scene may be generated according to the current scene and fed back to a user, instead of feeding back a single fixed telephone operation for all scenes.

The plurality of types of the recommendation responses are at least two types, the recommendation type corresponding to the user response of the current round can be determined, and then the recommendation response of the corresponding type is generated based on the recommendation type.

Specifically, multiple types can be preconfigured in the system, and when determining to trigger the dialog recommendation, the system can determine a recommendation type corresponding to the current round of user responses from among the multiple recommendation types and then generate a recommendation response corresponding to the recommendation type.

In the embodiment, the recommendation type corresponding to the user response is determined in the multiple recommendation types, so that the selection in the multiple recommendation types can be realized, the method is suitable for the current scene, the corresponding recommendation response is generated based on the recommendation type, the more flexible recommendation response can be generated, and the method is better suitable for the current scene. Further, the current recommendation response is generated based on the current user response, and real-time recommendation can be achieved.

Fig. 2 is a schematic diagram according to a second embodiment of the present disclosure. The embodiment provides a dialogue method, which comprises the following steps:

201. a task configuration is received.

The system developer can configure the task of the system, the task configuration refers to configuring the configuration data of the task corresponding to the current system for the system, the capability of the current system is defined, and different data in the configuration data can be used in the relevant links of task type man-machine conversation. The configuration data may include the following data:

(1) intention is: the purpose that the user wants to express in the conversation is that the whole conversation system is to help the user to fulfill the specific task requirements, such as movie ticket reservation, air ticket reservation, music playing and the like. A task-based dialog system may contain a number of intents, each of which includes: intent names, intent descriptions, intent example interrogatories, intent satisfiacs, etc.

(2) Word groove: key information required for achieving the purpose (intention) of a user conversation, such as information of a departure place, a destination, departure time and the like under the reservation intention of an air ticket, is a word slot. An intent may contain 0 or more word slots, the configuration under each word slot including: word slot names, word slot descriptions, word slot example values, word slot clarification techniques, etc.

(3) A question-answer library: background knowledge of a task-based dialog scene is used for solving the information query requirement of a user, and exists in a dialog form of question-answer pairs (usually one question-answer or multiple questions and multiple answers, that is, multiple similar questions correspond to multiple answers). The configuration of each question-answer pair comprises: questions and corresponding answers.

(4) Other configuration data: the system also comprises a conversation sample, a conversation template, a characteristic word and the like, is mainly used for improving the conversation understanding (or spoken language understanding) effect, and can also be used for a conversation recommendation process. The dialogue sample refers to historical dialogue information, the dialogue template refers to a template of the same sentence pattern obtained based on a plurality of pieces of dialogue information, and the feature words are words to be analyzed, such as the names of word slots.

202. A current user response to the user input is received.

203. And executing a specific task corresponding to the current user response based on the current user response.

As shown in fig. 3, when the system executes a specific task corresponding to a current user response, the system may perform processing such as Spoken Language Understanding (SLU), speech State Tracking (DST), Dialog Policy Learning (DPL), Natural Language Generation (NLG), and further, if the system includes a question-answer library, it may also perform query-library search and sorting based on the user response.

An SLU: the semantic understanding is performed on the user response to see whether a certain intention can be hit or not, and the included word slot information and the like.

DST: and generating the dialog state of the task-type multi-turn dialog according to the SLU understanding result of the current turn and the historical dialog state.

DPL: and deciding the reply action of the current round according to the result output by the DST, such as word slot clarification, intention satisfaction, conversation failure and the like. In addition, some task-based dialog scenes contain a question-answer library, and the question-answer library is searched and ranked at this time, so that answers corresponding to the hit questions should be output in the current round or word slot clarification is comprehensively decided according to matching results of the question-answer library and DST output results, and the like.

NLG: a natural language sentence is generated according to the action output by the DPL, and at this time, a task-based dialog response (or called task response) is obtained.

204. And judging whether to trigger conversation recommendation or not based on the execution result of the specific task, if so, executing 205, and otherwise, executing 208.

As shown in fig. 3, a Conversation Recommendation (CR) module may be included in the system, and the CR module may further include: a CR determiner, a CR classifier, and a CR executor.

Wherein the CR determiner may determine whether to trigger the dialog recommendation based on an execution result of the specific task.

The CR determiner may perform: if the execution result is that the specific task is successful and the specific task has a continuing task, determining that the execution result of the specific task triggers dialogue recommendation; or if the execution result is that the specific task fails, determining that the execution result of the specific task triggers dialog recommendation.

Further, the CR determiner may further perform: and if the execution result is that the specific task is successful and the specific task does not have a continuous task, determining that the execution result of the specific task does not trigger conversation recommendation, and feeding back a task response as a system response to the user by the system at the moment, namely, performing a normal task conversation process.

The CR arbiter mainly uses a rule-based method in the decision:

(1) task success and hit guidance rules: the execution result of a specific task is that the intention is satisfied (or called task success), and a continuation intention guiding rule defined in the task configuration is hit (when one intention is satisfied, the guidance is directly to another intention) (or called continuation task exists), and at this time, conversation recommendation is needed.

(2) User response failure to understand results in a dialog failure: when the user response is not understood, the task type dialog response can be caused to use the failure bottom dialect, or the word groove is caused to be clarified continuously, and the like, namely, the task fails at the moment, and the dialog recommendation is required at the moment.

When the conversation recommendation judgment result is ' yes ' (Y) ', entering a CR classifier to perform subsequent strategy processing; when the determination result is "no (N)", a task response is obtained and output as a system response. The task response can be obtained in a normal task conversation process, and details are not described.

By determining whether or not to perform a dialogue recommendation based on the execution result, it is possible to perform a dialogue recommendation when a dialogue recommendation is required and perform a normal task dialogue when a dialogue recommendation is not required.

205. And determining the current recommendation type corresponding to the current user response.

As shown in fig. 3, after the CR determiner determines that the dialog recommendation is triggered, the CR classifier may be triggered to execute, and the CR classifier may execute: and classifying the dialog information including the current user response by adopting a pre-trained classification model so as to determine the current recommendation type corresponding to the current user response.

Dialog information may also be referred to as dialog context (context), and generally speaking, a dialog is a plurality of rounds, and the dialog information may include user responses of a current round, system responses of the current round (if any), user responses of previous rounds, system responses of previous rounds, dialog states of the current round (such as intention, word slot, etc.), and so on.

The input of the CR classifier is the above-mentioned dialog information, and the output is the classification result, that is, which of the plurality of recommendation types the current recommendation type is.

The classification model can be trained in advance, can be a deep neural network model, and can be trained in a machine learning mode. Additionally, the classification model may also be iteratively updated.

For example, a small amount of training data needs to be constructed manually initially, a model is trained by taking dialogue information samples such as user response, task-based dialogue response, on-wheel system response, current-wheel dialogue state and the like as input, and a multi-classification result is output. Machine learning is a continuous iterative optimization process, after an initial model is put into use and a certain amount of dialogue logs are accumulated, more data are extracted from the dialogue logs to retrain the model, and the model achieves a better effect through repeated iteration.

Taking 5 recommended types as examples, the following may be specific:

(1) continuation task recommendation (or called continuation intention guide): when one intent is satisfied, the topic is directed to another intent. Not all dialog scenarios have such a continuing relationship between intentions, and need to be explicitly specified in terms of scenario characteristics when developing a dialog system.

(2) And (3) recommending system capacity: this generally occurs in two cases:

in the first case, the user actively asks for system capabilities: the user does not know the capability boundaries of the current dialog system and actively makes inquiries.

In the second case, the dialog fails during the non-task progress: the non-task-in-progress process means that the current round is not completing a specified task, and usually happens to cause dialog failure before the start of the task and dialog failure caused by the failure of the SLU after the completion of a certain task.

(3) And (3) slot filling conversation recommendation: the achievement of some tasks requires collecting some key information such as word groove, and at the moment, word groove missing clarification is carried out. When a word slot is not filled all the time, the system will make repeated queries, and this time will trigger a slot filling dialogue recommendation.

(4) Affirmative no dialog recommendation: some tasks may show the collected information for the user to confirm, for example: and confirming the order information when the train ticket is ordered, wherein the affirmation is that the attitude is clear. When no positive attitude is collected all the time, the system makes repeated inquiry, and at the moment, positive dialogue recommendation is triggered.

(5) Selecting a dialog recommendation: several options are presented for the user to select among certain tasks, such as: train number selection when booking train tickets, which is the clarification of multiple result selection. When the selection result is not collected all the time, the system makes repeated inquiry, and the selection dialogue recommendation is triggered at the moment.

Therefore, the CR classifier may employ a classification model, and based on the dialog information, may determine the current recommendation type among the 5 recommendation types. For example, if the task corresponding to the current user response is a train ticket booking and the task is successful, and the continuing task for booking the train ticket is configured in advance to check the weather, the current recommended type is a continuing task type, and then the recommended response corresponding to the continuing task can be generated.

The method and the device can determine the current recommended type by adopting the pre-trained classification model, and can improve the accuracy of the current recommended type because the classification model can be continuously updated.

206. And generating a current recommendation response corresponding to the current recommendation type based on the current recommendation type.

As shown in fig. 3, after the CR classifier determines the current recommendation type, the CR executor may be triggered to execute, and the CR executor may generate a corresponding current recommendation response based on the current recommendation type determined by the CR classifier.

207. And feeding back the current recommendation response to the user.

As shown in fig. 3, at the time of dialog recommendation, a recommendation response may be fed back to the user as a system response.

208. And obtaining the current task response and feeding back the current task response to the user.

If the dialog recommendation is not triggered, a normal task dialog process may be performed, for example, if the CR determiner determines that the dialog recommendation is not triggered, the system may generate a task response, and then, the task response may be fed back as a system response to the user.

Specifically, for example, if the execution result of the specific task is that the specific task is successful and there is no continuing task for the specific task, the dialog recommendation may not be triggered at this time, and a normal task response may be generated, such as "several cups are needed" in the above example.

When generating the current recommendation content based on the current recommendation type, the following method may be adopted: and generating a current recommendation response corresponding to the current recommendation type based on the recommendation template corresponding to the current recommendation type and/or including the dialog information responded by the current user.

When the task is configured, recommendation templates corresponding to various recommendation types in multiple recommendation types can be configured, the recommendation templates are general templates of various dialog information under related scenes, then, a current recommendation response is generated based on the general recommendation templates, for example, when the current recommendation type is a positive or negative dialog recommendation, the corresponding recommendation template can be configured in advance to be ' i don't understand your meaning, for example, yes or no ', and when the current recommendation type is determined to be a positive or negative dialog recommendation, the recommendation template can be fed back to the user as the current recommendation response. Further, corresponding to the dialog information of the current scene, all or part of the dialog information in the current scene may be used to perform processing such as filling, filtering, and the like on the recommendation template, so as to generate a current recommendation response of the current scene. For example, if the task of continuing to order a train ticket is to check the weather, the recommendation template corresponding to the task of continuing may be: weather that needs to query XX? XX in the list can be filled according to the dialogue information of the current scene, for example, when the destination slot obtained when the train ticket is ordered is shanghai, XX is shanghai. For another example, when the user inquires about the system capability, for example, the recommendation template corresponding to the system capability is "i will order a train ticket and order a drink", if the user has successfully completed the task of ordering a train ticket before, the train ticket may be deleted from the recommendation template, and "i will order a drink" is fed back to the user as the current recommendation response.

By generating the current recommendation response based on the recommendation template and/or the dialogue information, the recommendation response more suitable for the current scene can be generated, and the accuracy of the current recommendation response is improved.

Specifically, the method may include: if the current recommendation type is a new guide task type, acquiring a recommendation template corresponding to the new guide task type, processing the recommendation template corresponding to the new guide task type by adopting the dialogue information to obtain a processed recommendation template, and taking the processed recommendation template as the current recommendation response; or if the current recommendation type is the current task completion guidance type, acquiring a recommendation template corresponding to the current task completion guidance type, and taking the recommendation template corresponding to the current task completion guidance type as the current recommendation response.

Among the 5 recommendation types, the continuous task recommendation and the system capability recommendation can be regarded as guiding a new task type, and at this time, a current recommendation response can be obtained based on a corresponding recommendation template and dialog information; the slot filling conversation recommendation, the affirmative or negative conversation recommendation and the selective conversation recommendation can be performed by guiding the current task to complete the type, and at the moment, the current recommendation response can be obtained based on the corresponding recommendation template.

Through the different processing, the user can be quickly guided to carry out a new task or the user can be guided to complete the current task.

Further, the current recommendation responses corresponding to the above 5 recommendation types may be as follows:

(1) continuation task recommendation (or called continuation intention guide).

In the first case, the continuation-intention-leading rule has been explicitly defined in the task configuration: matching in the continuing intention guide rule of the task configuration by using the intention met by the current round, finding the next intention which should be guided to next, and generating the guide dialect according to the intention example question method and the dialogue state.

In the second case, no continuation intention guidance rule is defined in the task configuration: the method needs to perform association intention mining according to the existing dialog logs in the current scene, that is, when one intention is satisfied, most users tend to inquire which intention/intentions are next, obtain the association relation between the intentions by a method of mining logs, so as to form a continuing intention guide rule (if existing), and then generate a guide operation according to the guide rule.

The current recommendation response corresponding to the follow-up task recommendation can be as shown in fig. 4.

(2) System capability recommendation

In the first case, the user actively asks for system capabilities: and selecting the high-frequency intention to be displayed to the user, so that the user can know the capability of the system. In addition, it should be noted that when a certain task is already completed in the group of dialogs, the completed task should be avoided when performing the system capability recommendation, so as to avoid repeated recommendations. As shown in fig. 5.

In the second case, the question-answer pair matching similarity is too low to cause dialog failure: and comprehensively considering the similarity and the occurrence frequency of question-answer pairs to recommend the system capacity. As shown in fig. 6.

In the third case, a session failure beyond the current system capability boundary: and carrying out capability recommendation of the current system. In addition, it should be noted that when a certain task is already completed in the group of dialogs, the completed task should be avoided when performing the system capability recommendation, so as to avoid repeated recommendations. As shown in fig. 7.

(3) Slot filling conversation recommendation

In the process of task achievement, a certain word slot is not filled all the time, and at this time, a user needs to be guided to enable the word slot to be filled, and in addition, an example word slot value can be shown in some cases to help the user to better understand, the example word slot value can be dynamically generated according to the user response (if any), and a static value can be obtained from task configuration. As shown in fig. 8.

(4) Kennel dialog recommendations

In the process of affirming, no affirmation attitude of the user is collected, and at this time, a certain guidance needs to be given to the user to prompt the expression of the affirmation attitude. As shown in fig. 9.

(5) Selecting conversation recommendations

In the process of multi-result selection, the selection results of the user are not collected all the time, and at this time, a certain guidance needs to be given to the user to prompt the user to select the listed options. As shown in fig. 10.

After the current recommendation response is obtained, the current recommendation response can be fed back to the user as a system response.

In the embodiment, the purpose of task achievement in the task type dialog system is always achieved, the dialog configuration is fully utilized, the dialog recommendation can be carried out in real time, and the task completion rate is well improved. By the aid of the conversation recommendation determiner, whether conversation recommendation needs to be used in the current round of conversation can be well determined, intervention is not performed in the process of normal running of tasks, and task type conversation responses are directly output; and when the conversation is abnormal, the conversation recommendation is triggered and a conversation recommendation response is output. All cases of session failures can be handled, whether before the task starts, during the task or after the task ends, the session recommendation is fully considered, and the recommendation types are very diverse. The conversation recommendation (such as the slot filling conversation recommendation, the affirmative conversation recommendation and the like) in the task process not only can reasonably guide the user to use the current conversation system in a standard way, but also can keep the current task to continue to promote the task to be achieved. The two aspects of intention recommendation and question and answer recommendation are comprehensively considered, and the conversation requirements under various scenes are met. The recommended dialect is smoothly linked with the user response, the continuity and the fluency of the dialect are ensured, and the man-machine conversation experience is real.

Fig. 11 is a schematic diagram of an eleventh embodiment according to the present disclosure, which provides a dialog device. As shown in fig. 11, the conversation device 1100 includes: an execution module 1101, a determination module 1102, and a recommendation module 1103.

The execution module 1101 is configured to execute a specific task corresponding to a current user response based on the current user response input by a user; the determining module 1102 is configured to determine, if the execution result of the specific task triggers a dialog recommendation, a current recommendation type corresponding to the current user response among multiple recommendation types; the recommending module 1103 is configured to generate a current recommendation response corresponding to the current recommendation type based on the current recommendation type, and feed the current recommendation response back to the user.

In some embodiments, the apparatus further comprises: the triggering module is used for determining that the execution result of the specific task triggers the dialogue recommendation if the execution result is that the specific task is successful and the specific task has a continuing task; or if the execution result is that the specific task fails, determining that the execution result of the specific task triggers dialog recommendation.

In some embodiments, the determining module 1102 is specifically configured to: and classifying the dialog information including the current user response by adopting a pre-trained classification model so as to determine the current recommendation type corresponding to the current user response.

In some embodiments, the recommending module 1103 is specifically configured to: and generating a current recommendation response corresponding to the current recommendation type based on the recommendation template corresponding to the current recommendation type and/or including the dialog information responded by the current user.

In some embodiments, the recommending module 1103 is further specifically configured to: if the current recommendation type is a new guide task type, acquiring a recommendation template corresponding to the new guide task type, processing the recommendation template corresponding to the new guide task type by adopting the dialogue information to obtain a processed recommendation template, and taking the processed recommendation template as the current recommendation response; or if the current recommendation type is the current task completion guidance type, acquiring a recommendation template corresponding to the current task completion guidance type, and taking the recommendation template corresponding to the current task completion guidance type as the current recommendation response.

In the embodiment of the disclosure, the recommendation type corresponding to the user response is determined in the multiple recommendation types, so that the selection in the multiple recommendation types can be realized, the method is suitable for the current scene, the corresponding recommendation response is generated based on the recommendation type, the more flexible recommendation response can be generated, and the method is better suitable for the current scene. Further, the current recommendation response is generated based on the current user response, and real-time recommendation can be achieved.

It is to be understood that in the disclosed embodiments, the same or similar elements in different embodiments may be referenced.

It is to be understood that "first", "second", and the like in the embodiments of the present disclosure are used for distinction only, and do not indicate the degree of importance, the order of timing, and the like.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 12 shows a schematic block diagram of an example electronic device 1200, which can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 12, the electronic apparatus 1200 includes a computing unit 1201, which can perform various appropriate actions and processes in accordance with a computer program stored in a Read Only Memory (ROM)1202 or a computer program loaded from a storage unit 1208 into a Random Access Memory (RAM) 1203. In the RAM 1203, various programs and data necessary for the operation of the electronic apparatus 1200 may also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other by a bus 1204. An input/output (I/O) interface 1205 is also connected to bus 1204.

Various components in the electronic device 1200 are connected to the I/O interface 1205, including: an input unit 1206 such as a keyboard, a mouse, or the like; an output unit 1207 such as various types of displays, speakers, and the like; a storage unit 1208, such as a magnetic disk, optical disk, or the like; and a communication unit 1209 such as a network card, modem, wireless communication transceiver, etc. The communication unit 1209 allows the electronic device 1200 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

The computing unit 1201 may be a variety of general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The computing unit 1201 executes the respective methods and processes described above, such as the dialogue method. For example, in some embodiments, the conversation understanding method or conversation method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1208. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 1200 via the ROM 1202 and/or the communication unit 1209. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the dialog method described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to perform the dialog method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The Server can be a cloud Server, also called a cloud computing Server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server may also be a server of a distributed system, or a server incorporating a blockchain.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and the present disclosure is not limited herein.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.