1. A title approval method is characterized by comprising the following steps:

identifying all characters of the image to be corrected, and distinguishing printed characters from handwritten characters;

segmenting the big questions of the image to be corrected, and aiming at each big question, carrying out the following processes:

obtaining the question type of the big question, and obtaining a question type structure corresponding to the big question according to the question type of the big question; and

according to the question type structure corresponding to the big question, splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively, and forming at least one correcting unit, wherein the correcting unit comprises at least one answer character and at least one question stem part;

and retrieving and correcting each correcting unit based on the spliced correcting units.

2. The method according to claim 1, wherein the step of segmenting the image to be corrected into the big questions is realized by the question number and the line spacing; the line spacing is calculated through the position information of the identified printed characters;

optionally, the step of performing segmentation on the big topic of the image to be corrected further includes: and searching based on the identification characters of the big questions, and if a first question with the matching degree higher than a threshold value is searched in a database, realizing segmentation based on the comparison result of the big questions and the first question in the segmentation step.

3. The method of claim 1, wherein the step of retrieving and modifying based on the spliced modifying unit comprises:

retrieving based on the spliced correction units to obtain corresponding standard answers;

and judging whether the answer characters are correct or not based on the answer characters answered by the user, the question stem context, the question type structure, the answer category, the subject and the subject section and the standard answer.

4. The method according to claim 3, wherein the step of retrieving the corresponding standard answer based on the spliced modifying unit comprises:

converting each correcting unit into vector expression, and retrieving by using the characteristic vector to obtain a standard answer;

optionally, the step of converting the correction unit into a vector expression comprises:

splicing the question stem part and answer characters answered by the user according to the question types to form a text string, inputting the text string into a first artificial intelligent model through a character input dictionary, and converting the text string into vector expression;

optionally, the step of retrieving by using the feature vector to obtain the standard answer includes:

searching a corresponding standard answer in an answer database through the feature vector in the vector expression; the answer database stores structured standard answers corresponding to the question types and the question stem contents.

5. The method of claim 4, wherein the first artificial intelligence model is a Word2vec model, a textCNN model, or a BERT pre-training model.

6. The method according to any one of claims 3 or 4, wherein the step of determining whether the answer text is correct comprises:

inputting the question stem part, answer characters answered by the user and the obtained standard answers into a second artificial intelligence model trained in advance for secondary classification, and outputting a conclusion whether the answer characters are correct or not;

the second artificial intelligence model is a convolution neural network model.

7. The method of claim 1, wherein the step of distinguishing between printed text and handwritten text is performed by a third artificial intelligence model trained in advance;

optionally, the third artificial intelligence model is a convolutional neural network model for implementing classification;

optionally, the position information of the printed characters and the handwritten characters is recorded when the printed characters and the handwritten characters are recognized.

8. A topic modification system, comprising:

the character distinguishing module is used for identifying all characters in the image to be corrected and distinguishing the printed characters from the handwritten characters;

the big question segmentation module is used for segmenting the big questions of the image to be corrected;

the question type structure obtaining module is used for obtaining the question type of the big question and obtaining the question type structure corresponding to the big question according to the question type of the big question;

the splicing module is used for splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively according to the question type structure corresponding to the big question and forming at least one correcting unit, and the correcting unit comprises at least one answer character and at least one question stem part;

the retrieval verification answer module is used for retrieving based on the spliced correction units and verifying whether answer characters answered by the user are correct or not;

and the correcting module is used for correcting the correcting unit based on the verification result of the retrieval verification answer module.

9. An electronic device comprising a processor and a memory, the memory for storing a computer-executable program, characterized in that:

when the computer-executable program is executed by the processor, the processor performs the title wholesale method of any one of claims 1-7.

10. A computer-readable medium storing a computer-executable program, wherein the computer-executable program, when executed, implements the title revising method according to any one of claims 1-7.

Background

With the technical progress, the current students can shoot the difficult problems which can not be solved through a mobile phone to search answers through a network when encountering the difficult problems, or test questions handwritten by a sign pen or a ball pen are uploaded through shooting and corrected by a machine, so that the search for knowledge is greatly facilitated, and the shortage of teachers and materials is made up.

However, the machine is a machine and cannot understand and think the inherent logic and meaning of the input content like a human being, so that when a student shoots and uploads a large test question, the machine cannot understand the interrelation and specific meaning of each word, and a technician needs to carefully design the machine to accurately correct the student's answer. In addition, in the hand-written test questions which can be corrected by the machine, because the answers may have different expression modes, the answer database is not necessarily completely recorded, which may cause the answer answered by the user to be judged wrongly due to the deviation of one word, and the word does not affect the correctness of the answer, so how to accurately identify and correct the hand-written answer of the answerer, especially the answer which is similar to the standard answer but still correct, becomes a technical problem to be researched and solved urgently.

Disclosure of Invention

In view of the above, the present invention is directed to a topic modification method and system, and an electronic device and a computer readable medium using the same, which are intended to at least partially solve at least one of the above technical problems.

In order to achieve the above object, as a first aspect of the present invention, there is provided a title correction method, including the steps of:

identifying all characters of the image to be corrected, and distinguishing printed characters from handwritten characters;

segmenting the big questions of the image to be corrected, and aiming at each big question, carrying out the following processes:

obtaining the question type of the big question, and obtaining a question type structure corresponding to the big question according to the question type of the big question; and

according to the question type structure corresponding to the big question, splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively, and forming at least one correcting unit, wherein the correcting unit comprises at least one answer character and at least one question stem part;

and retrieving and correcting each correcting unit based on the spliced correcting units.

Optionally, the step of segmenting the image to be corrected into major questions is implemented by using question numbers and line intervals; the line spacing is calculated through the position information of the identified printed characters;

optionally, the step of performing segmentation on the big topic of the image to be corrected further includes: and searching based on the identification characters of the big questions, and if a first question with the matching degree higher than a threshold value is searched in a database, realizing segmentation based on the comparison result of the big questions and the first question in the segmentation step.

Optionally, the step of retrieving and modifying based on the spliced modifying unit includes:

retrieving based on the spliced correction units to obtain corresponding standard answers;

and judging whether the answer characters are correct or not based on the answer characters answered by the user, the question stem context, the question type structure, the answer category, the subject and the subject section and the standard answer.

Optionally, the step of retrieving and acquiring a corresponding standard answer based on the spliced modifying unit includes:

converting each correcting unit into vector expression, and retrieving by using the characteristic vector to obtain a standard answer; searching a corresponding standard answer in an answer database through the feature vector in the vector expression;

optionally, the step of converting the correction unit into a vector expression comprises:

and splicing the question stem part and the user answering part according to the question type to form a text string, and inputting the text string into a first artificial intelligent model through a character input dictionary to convert the text string into vector expression.

Optionally, the first artificial intelligence model is a Word2vec model, a textCNN model, or a BERT pre-training model.

Optionally, the step of determining whether the answer text is correct includes:

inputting the question stem part, answer characters answered by the user and the obtained standard answers into a second artificial intelligence model trained in advance for secondary classification, and outputting a conclusion whether the answer characters are correct or not;

the second artificial intelligence model is a convolution neural network model.

Optionally, the step of distinguishing the printed text from the handwritten text is implemented by a third artificial intelligence model trained in advance;

optionally, the third artificial intelligence model is a convolutional neural network model for implementing classification;

optionally, the position information of the printed characters and the handwritten characters is recorded when the printed characters and the handwritten characters are recognized.

As a second aspect of the present invention, there is also provided a title correction system, including:

the character distinguishing module is used for identifying all characters in the image to be corrected and distinguishing the printed characters from the handwritten characters;

the big question segmentation module is used for segmenting the big questions of the image to be corrected;

the question type structure obtaining module is used for obtaining the question type of the big question and obtaining the question type structure corresponding to the big question according to the question type of the big question;

the splicing module is used for splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively according to the question type structure corresponding to the big question and forming at least one correcting unit, and the correcting unit comprises at least one answer character and at least one question stem part;

the retrieval verification answer module is used for retrieving based on the spliced correction units and verifying whether answer characters answered by the user are correct or not;

and the correcting module is used for correcting the correcting unit based on the verification result of the retrieval verification answer module.

As a third aspect of the present invention, there is also provided an electronic device, including a processor and a memory, the memory storing a computer-executable program, when the computer-executable program is executed by the processor, the processor executing the title correcting method as described above.

As a fourth aspect of the present invention, there is also provided a computer-readable medium storing a computer-executable program which, when executed, implements the title correction method as described above.

Based on the above technical solution, the title correction method and system of the present invention have at least one of the following advantages compared with the prior art:

the invention systematically provides the method steps of identifying various elements in the image to be identified, distinguishing and splicing the elements into a plurality of correction units, thereby accurately identifying the question stem information and the user answering part under various question types;

the invention can combine the user answering part and the context thereof, convert the user answering part into a vector through a character input dictionary, and improve the retrieval efficiency by using the characteristic vector representation;

the invention identifies whether the answering part of the user is the same as or similar to the standard answer or not through the first and second classification models, and whether the answering part of the user can also be a correct answer, thereby expanding the tolerance of answering and avoiding the condition of dead set of standard answers which is too rigid during correction.

Drawings

FIG. 1 is a block flow diagram of a topic modification method of the present invention;

FIG. 2 is a schematic diagram of a topic modification system according to the present invention;

FIG. 3 is a schematic diagram of the electronic device of the present invention;

FIG. 4 is a schematic diagram of a computer readable medium of the present invention;

FIG. 5 is a photograph showing the effect of practical processing in embodiment 1 of the present invention;

fig. 6 is a photograph showing another practical effect of the treatment according to embodiment 1 of the present invention.

Detailed Description

In describing particular embodiments, specific details of structures, properties, effects, or other features are set forth in order to provide a thorough understanding of the embodiments by one skilled in the art. However, it is not excluded that a person skilled in the art may implement the invention in a specific case without the above-described structures, performances, effects or other features.

The flow chart in the drawings is only an exemplary flow demonstration, and does not represent that all the contents, operations and steps in the flow chart are necessarily included in the scheme of the invention, nor does it represent that the execution is necessarily performed in the order shown in the drawings. For example, some operations/steps in the flowcharts may be divided, some operations/steps may be combined or partially combined, and the like, and the execution order shown in the flowcharts may be changed according to actual situations without departing from the gist of the present invention.

The block diagrams in the figures generally represent functional entities and do not necessarily correspond to physically separate entities. I.e. these functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and/or processing module means and/or microcontroller means.

The same reference numerals denote the same or similar elements, components, or parts throughout the drawings, and thus, a repetitive description thereof may be omitted hereinafter. It will be further understood that, although the terms first, second, third, etc. may be used herein to describe various elements, components, or sections, these elements, components, or sections should not be limited by these terms. That is, these phrases are used only to distinguish one from another. For example, a first device may also be referred to as a second device without departing from the spirit of the present invention. Furthermore, the term "and/or", "and/or" is intended to include all combinations of any one or more of the listed items.

On one hand, the system provides the method steps of identifying various elements in the image to be identified, distinguishing and splicing the elements into a plurality of correction units, so that the question stem information and the user response under various question types can be accurately identified, and the expansibility is good.

The invention is provided aiming at the situation that the existing answer is required to be completely consistent with the standard answer sometimes and slightly deformed even though the answer is wrong, can avoid too stiff in the correction process and can accurately and efficiently complete the correction task of approximate answers.

It should be noted that the present invention is not only suitable for gap filling, but also can be used as long as the corresponding question type structure can be obtained and the contents can be accurately spliced. In addition, for the blank filling question, the blank filling question is not only called as a question to be answered on a lower line or in parentheses, and the blank filling question can be regarded as a blank filling question such as a large small question, a sequence of questions, a number of special questions and a idiom connecting question which are represented by a circle, an oval circle, a square grid and the like and need to be filled with contents in the middle. The invention is not only suitable for correcting the mathematics problem but also can correct various subjects such as language, history, geography, … … and the like from the whole algorithm.

Based on this, as shown in fig. 1, the present invention provides a title modifying method, which comprises the following steps:

identifying all characters of the image to be corrected, and distinguishing printed characters from handwritten characters;

segmenting the big questions of the image to be corrected, and aiming at each big question, carrying out the following processes:

obtaining the question type of the big question, and obtaining a question type structure corresponding to the big question according to the question type of the big question; and

according to the question type structure corresponding to the big question, splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively, and forming at least one correcting unit, wherein the correcting unit comprises at least one answer character and at least one question stem part;

and retrieving and correcting each correcting unit based on the spliced correcting units.

The term "topic" is a concept defined in the present invention, and refers to a topic or a group of topics of the same topic type, which are usually distinguished by a topic number. A "large topic" may include multiple small topics with labels, or multiple unnumbered topic types with similar structures. The specific example is shown in fig. 5 and 6, the first major topic in fig. 5 is a choice initial or final topic, and includes 6 unnumbered topics with similar structures, i.e., 6 modification modules; the second main topic in fig. 5 is a pinyin-writing word topic, which includes 4 unnumbered topics with similar structures, i.e., 4 correction modules; the third topic in FIG. 6 includes 5 topics, each of which is considered as a modification module.

Wherein, the step of segmenting the image to be corrected into big questions is realized by the question number or the line spacing; the question number can be directly identified by identifying the first line of the question characters, such as ' one ', ' two ', ' or ' 5 ', and the like, the line spacing is calculated by the position information of the identified printed characters, for example, in some cases, the line spacing in the big questions is 1.5 times, the line spacing between the big questions is 2.0 times, and the difference of the line spacing is a way of distinguishing different questions.

Optionally, when the image to be modified is subjected to the segmentation of the big topic, the auxiliary confirmation standard may also be utilized: and searching based on the recognized characters of the big questions, wherein if the first question with the matching degree higher than the threshold value is searched in the database, the segmentation step can realize segmentation based on the comparison result of the big questions and the first question, in other words, if the big questions and the first question are found in the database, the part which is the same as the big questions is the big questions, and the different part can be eliminated.

Before the splicing step, it is preferable to identify redundant characters, for example, by training a model or determining whether the answer of the user is an answer character or a redundant character according to whether the handwriting position falls into a specific area or not. The recognition of the redundant text can be performed based on the position information or by training a model, which is specifically described in detail in another patent application and is not described herein again.

The step of retrieving and correcting based on the spliced correcting units comprises the following steps:

retrieving based on the spliced correction units to obtain corresponding standard answers;

and judging whether the answer words are correct or not based on the answer words answered by the user, the question stem context, the question type structure, the answer category, the subject and the subject section, the standard answer and the like.

The step of retrieving and acquiring the corresponding standard answer based on the spliced correction unit comprises the following steps:

converting each correcting unit into vector expression, and retrieving by using the characteristic vector to obtain a standard answer;

optionally, the step of converting the correction unit into a vector expression comprises:

and splicing the question stem part and answer characters answered by the user according to the question type to form a text string, inputting the text string into a first artificial intelligent model (embedding) through a character input dictionary, and converting the text string into vector expression. The question stem part can be divided into separate question stem parts (judgment questions and selection questions) according to the question types, and the question stem parts are divided into an upper part and a lower part (blank filling questions with blank filling positions in the middle) and the like, and the complete context text string is actually formed by splicing the answer characters answered by the user and then converted into vector expression.

Optionally, the step of retrieving by using the feature vector to obtain the standard answer includes:

searching a corresponding standard answer in an answer database through the feature vector in the vector expression; the answer database stores structured standard answers corresponding to the question types and the question stem contents, for example, a blank filling question comprises a question stem part, a question blank setting position and the standard answers.

The first artificial intelligence model is, for example, a Word2vec model, a textCNN model, or a BERT pre-training model, and is preferably a Word2vec model.

Optionally, the step of determining whether the answer text is correct includes:

inputting the question stem part, answer characters answered by the user and the obtained standard answers into a second artificial intelligence model trained in advance for secondary classification, and outputting a conclusion whether the answer characters are correct or not;

the second artificial intelligence model is, for example, a neural network model, preferably a Convolutional Neural Network (CNN) model, which mainly performs a two-class classification, so that any classification model capable of searching for an approximate answer based on a standard answer may be used. The second artificial intelligence model can be obtained by training a large number of samples from actual Query on a platform of the company, so that the accuracy of identification can be higher due to the closer proximity to a real environment.

Optionally, the step of distinguishing the printed text from the handwritten text is implemented by a third artificial intelligence model trained in advance;

optionally, the third artificial intelligence model is, for example, a Convolutional Neural Network (CNN) model implementing classification.

Optionally, the position information of the printed characters and the handwritten characters is recorded when the printed characters and the handwritten characters are recognized.

Optionally, in the step of modifying the answer text, a modification mark with a different color, such as "v" or "x" or "good" may be provided at a set distance in a set direction of the "answer text" based on the previous determination result, and the specific arrangement is detailed in another patent application, which is not described herein again.

As shown in FIG. 2, the present invention further provides a title correction system, comprising:

the character distinguishing module is used for identifying all characters in the image to be corrected and distinguishing the printed characters from the handwritten characters;

the big question segmentation module is used for segmenting the big questions of the image to be corrected;

the question type structure obtaining module is used for obtaining the question type of the big question and obtaining the question type structure corresponding to the big question according to the question type of the big question;

the splicing module is used for splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively according to the question type structure corresponding to the big question and forming at least one correcting unit, and the correcting unit comprises at least one answer character and at least one question stem part;

the retrieval verification answer module is used for retrieving based on the spliced correction units and verifying whether answer characters answered by the user are correct or not;

and the correcting module is used for correcting the correcting unit based on the verification result of the retrieval verification answer module.

The big question cutting module is used for cutting the big questions of the image to be corrected according to the question numbers or line intervals; the question number can be directly identified by identifying the first line of the question characters, such as ' one ', ' two ', ' or ' 5 ', and the like, the line spacing is calculated by the position information of the identified printed characters, for example, in some cases, the line spacing in the big questions is 1.5 times, the line spacing between the big questions is 2.0 times, and the difference of the line spacing is a way of distinguishing different questions.

Optionally, when the segmentation module performs segmentation on the big topic of the image to be corrected, the auxiliary confirmation standard may also be used: and searching based on the recognized characters of the big problems, wherein if a first problem with a matching degree higher than a threshold value is searched in the database, the segmentation step can realize segmentation based on the comparison result of the big problems and the first problem, in other words, if the big problems are found in the database, the parts which are the same as the big problems are the big problems, and the different parts are eliminated.

The system further comprises a redundant character recognition module for preferably recognizing the redundant characters before the splicing step, for example, judging whether the answer of the user is an answer character or the redundant characters through a training model or according to whether the handwriting position falls into a specific area or not. The recognition of the redundant text can be performed based on the position information or by training a model, which is specifically described in detail in another patent application and is not described herein again.

Wherein, when the retrieval verification answer module is used for retrieving and correcting based on the spliced correction units, the retrieval verification answer module comprises:

retrieving based on the spliced correction units to obtain corresponding standard answers;

and judging whether the answer words are correct or not based on the answer words answered by the user, the question stem context, the question type structure, the answer category, the subject and the subject section, the standard answer and the like.

Wherein, when the retrieval verification answer module retrieves and acquires the corresponding standard answer based on the spliced correction unit, the retrieval verification answer module comprises:

converting each correcting unit into vector expression, and retrieving by using the characteristic vector to obtain a standard answer;

optionally, the step of converting the wholesale module into a vector expression comprises:

and splicing the question stem part in the correcting module and answer characters answered by the user into a character string in sequence based on the question types, inputting the character string into a first artificial intelligence model (embedding) through a character input dictionary, and converting the character string into vector expression. The question stem part can be divided into independent question stem parts (judgment questions and selection questions) or two parts of the upper and the lower (blank filling questions with blank filling positions in the middle) and the like based on the question types, and the complete context text string is actually formed by splicing the question stem parts with answer characters answered by the user and then converted into vector expression.

Optionally, the step of retrieving by using the feature vector to obtain the standard answer includes:

searching a corresponding standard answer in an answer database through the feature vector in the vector expression; the answer database stores structured standard answers corresponding to the question types and the question stem contents.

The first artificial intelligence model is, for example, a Word2vec model, a textCNN model, or a BERT pre-training model, and is preferably a Word2vec model.

The answer database stores formatted standard answers, such as a blank filling question including a question stem part, a question blank setting position and the standard answers.

Optionally, the step of determining whether the answer text is correct includes:

inputting the question stem part, answer characters answered by the user and the obtained standard answers into a second artificial intelligence model trained in advance for secondary classification, and outputting a conclusion whether the answer characters are correct or not;

the second artificial intelligence model is, for example, a neural network model, preferably a Convolutional Neural Network (CNN) model, which mainly performs a two-class classification, so that any classification model capable of searching for an approximate answer based on a standard answer may be used. The second artificial intelligence model can be obtained by training a large number of samples from actual Query on a platform of the company, so that the accuracy of identification can be higher due to the closer proximity to a real environment.

Optionally, the step of distinguishing the printed text from the handwritten text is implemented by a third artificial intelligence model trained in advance;

optionally, the third artificial intelligence model is, for example, a Convolutional Neural Network (CNN) model implementing classification.

Optionally, the position information of the printed characters and the handwritten characters is recorded when the printed characters and the handwritten characters are recognized.

Optionally, in the step of modifying the answer text, a modification mark with a different color, such as "check mark" or "x", may be provided at a set distance of a set orientation of the "answer text" based on the previous determination result, and the specific setting is detailed in another patent application, which is not described herein again.

The invention also discloses an electronic device, fig. 3 is a schematic structural diagram of the electronic device of the invention, and as shown in fig. 3, the electronic device includes a processor and a memory, the memory is used for storing a computer executable program, wherein when the computer executable program is executed by the processor, the processor executes the method as described above.

The electronic device may be embodied in the form of a general purpose computing device, for example. The number of the processors may be one, or may be multiple and work together. The invention also does not exclude that distributed processing is performed, i.e. the processors may be distributed over different physical devices. The electronic device of the present invention is not limited to a single entity, and may be a sum of a plurality of entity devices.

In which a memory stores a computer-executable program, typically machine-readable code, which is executable by the processor to enable an electronic device to perform the method of the invention, or at least some of the steps of the method.

The memory may include volatile memory, such as Random Access Memory (RAM) and/or cache memory, and non-volatile memory, such as read-only memory (ROM).

Optionally, in this embodiment, the electronic device further includes an I/O interface, which is used for data exchange between the electronic device and an external device. The I/O interface may be a local bus representing one or more of several types of bus structures, including a memory module bus or memory module controller, a peripheral bus, an accelerated graphics port, a processing module, or any other type of bus structure.

Elements or components not shown in the above examples may also be included in the electronic device of the present invention. For example, some electronic devices further include a display module such as a display screen, and some electronic devices further include a human-computer interaction element such as a button, a keyboard, and the like. Electronic devices are considered to be covered by the present invention as long as the electronic devices are capable of executing a computer-readable program in a memory to implement the method of the present invention or at least a part of the steps of the method.

Fig. 4 is a schematic diagram of a computer-readable recording medium of the present invention, as shown in fig. 4, on which a computer-executable program is stored, wherein the computer-executable program, when executed, implements the method as described above.

A computer readable storage medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable storage medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Python, Java, C + + or the like and conventional procedural programming languages, such as the C language, assembly language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

In order that the objects, technical solutions and advantages of the present invention will become more apparent, the present invention will be further described in detail with reference to the accompanying drawings in conjunction with the following specific embodiments. It should be noted that the following examples are only for illustrating the present invention and are not intended to limit the present invention.

Example 1

The title approval system of this embodiment is loaded in the memory of the mobile phone, and includes:

the character distinguishing module is used for identifying all characters in the image to be corrected and distinguishing the printed characters from the handwritten characters;

the big question segmentation module is used for segmenting the big questions of the image to be corrected;

the question type structure obtaining module is used for obtaining the question type of the big question and obtaining the question type structure corresponding to the big question according to the question type of the big question;

the splicing module is used for splicing the printed characters and the handwritten characters into a question stem part, answer characters answered by a user and redundant characters respectively according to the question type structure corresponding to the big question and forming at least one correction unit, and the correction unit comprises at least one answer character and at least one question stem part;

the retrieval verification answer module is used for retrieving and verifying whether answer characters answered by the user are correct or not based on the spliced correction units, and specifically comprises the following steps:

and splicing the question stem part in each correcting unit and answer characters answered by the user into a character string in sequence based on the question types, inputting the character string into a first artificial intelligence model (embedding) through a character input dictionary, and converting the character string into vector expression.

Inputting the question stem part, answer words answered by the user and the obtained standard answers in each correction unit into a second artificial intelligence model trained in advance for secondary classification, and outputting a conclusion whether the answer words are correct or not;

the first artificial intelligence model is a Word2vec model, formatted standard answers are stored in the answer database, the second artificial intelligence model is a Convolutional Neural Network (CNN) model, training samples of the second artificial intelligence model come from actual Query on a platform of the company, and therefore the accuracy of identification can be higher due to the fact that the second artificial intelligence model is closer to a real environment.

And the correcting module is used for correcting the correcting unit based on the verification result of the retrieval verification answer module.

Fig. 5 is a photograph showing the effect of actual processing in example 1 of the present invention. As shown in fig. 5, four question types appear in fig. 5, the system firstly performs segmentation of a large question dimension, five segmentation frames appear in a page, the first segmentation frame is a blank frame (which may be image residual information and the like to affect the recognition result), evaluation is not affected, and the other four segmentation frames are all accurate. Secondly, the system obtains the question type structure, the first big question is supplementary pinyin, and the pinyin belonging to the answer and the characters of the question stem are in the front and back structures; the second main question is the situation that Chinese characters are written by looking at pinyin, and the pinyin belonging to the question stem and the Chinese characters belonging to the answer belong to the upper and lower structures; the third main question is word selection and space filling, namely, firstly identifying that the printing font in the upper square box is a to-be-selected item (i-ninu), and then judging whether each bracket is a correct answer; the fourth topic is a bracket filling the blank, and a plurality of blank topics (the specific topic configuration of which is recorded in advance) indicated by the bracket are required to be identified. Then, the system needs to search the structured standard answers from the answer database according to the identified related information in the image, verify whether the answers of the user are correct, and carry out correction in sequence. From the results of drawing √ and mistaking the drawing √ good, the system also accurately identifies the corresponding question type structure and gives accurate judgment.

Fig. 6 is another photograph showing the effect of actual processing in embodiment 1 of the present invention. As shown in FIG. 6, the third big topic can be accurately divided, the 4 th, 5 th and 6 th small topics of the second big topic are divided into two topic types due to large form difference, the final judgment result is not affected, the fourth big topic is too disordered and does not form a division frame, but the small topics are corrected.

The second question of the fourth topic is vertical calculation, and it can be seen that the scroll face includes a miscalculated part, a carry flag, and a final answer in addition to the vertical form of the calculation. The system judges the answer characters through the redundant character recognition model, then searches for standard answers based on the question stem content, then verifies whether the answer characters are correct answers through the second model, if yes, marks a square root behind the answer characters, and if not, the answer characters are drawn to be good.

Therefore, the method has strong fault-tolerant capability, can adapt to various complicated question types, and accurately revises the handwritten answers.

While the foregoing embodiments have described the objects, aspects and advantages of the present invention in further detail, it should be understood that the present invention is not inherently related to any particular computer, virtual machine or electronic device, and various general-purpose machines may be used to implement the present invention. The invention is not to be considered as limited to the specific embodiments thereof, but is to be understood as being modified in all respects, all changes and equivalents that come within the spirit and scope of the invention.