Automatic Problem-Solving Method, Device and Electronic Device for Selecting Test Questions

Abstract

The present invention discloses an automatic problem-solving method, device and electronic device for multiple-choice questions. The automatic problem-solving method for multiple-choice questions includes: obtaining the text structured data of the multiple-choice questions; obtaining the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data; and generating the final automatic problem-solving answers by integrating the answers of the deep learning classification algorithm and the answers of the strategy algorithm. The present invention adopts a combination of a deep learning classification algorithm and a rule strategy algorithm for the automatic solution of multiple-choice questions, and comprehensively generates the final automatic problem-solving answers to ensure the usability and accuracy of the automatic solution of multiple-choice questions.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and particularly to an automatic problem-solving method, device and electronic device for selecting test questions. Background Art

[0002] Natural language generally refers to a language that naturally evolves with culture. For example, English, Chinese, and Japanese are examples of natural languages. Natural language is the main tool for human communication and thinking, and it is the crystallization of human wisdom. Therefore, in China, from primary school to high school, the learning and mastery of natural language are continuous.

[0003] In order to help primary and middle school students better learn and master at least one natural language, a large number of multiple-choice questions are included in the training test questions for primary and middle schools. Taking the study of English as an example, English - as a natural language, is the most widely studied second language and is one of the official languages or an official language in nearly 60 sovereign countries. See Figure 1 An example of an English multiple-choice question shown includes a text part and alternative options. The text part is missing some words, and the text part is completed by selecting the correct answer from the alternative options.

[0004] For this type of multiple-choice question as shown in Figure 1 it cannot be automatically answered. When students answer multiple-choice questions, those without a certain level of knowledge cannot provide tutoring. In addition, for multiple-choice questions, since they cannot be automatically answered, automatic marking cannot be achieved based on automatic answering; only the matching method based on pre-entered standard reference answers can be used to achieve automatic marking. This method increases the workload of entering standard reference answers and has high requirements for the source of multiple-choice questions, resulting in a poor experience for teachers in using automatic marking.

[0005] In view of this, the present invention patent is specifically proposed. Summary of the Invention

[0006] To solve the above problems, the present invention provides an automatic problem-solving method, device and electronic device for multiple-choice questions. Specifically, the following technical solutions are adopted:

[0007] An automatic problem-solving method for multiple-choice questions includes:

[0008] Obtaining the text structured data of the multiple-choice question;

[0009] Obtaining the answer of the deep learning classification algorithm and the answer of the strategy algorithm based on the text structured data;

[0010] Integrating the answer of the deep learning classification algorithm and the answer of the strategy algorithm to generate the final automatic problem-solving answer.

[0011] As an alternative embodiment of the present invention, obtaining the answer of the deep learning classification algorithm and the answer of the policy algorithm based on the text structured data includes:

[0012] Perform corresponding data processing on the text structured data to generate unified standard structured data;

[0013] Select the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected questions through the deep learning classification algorithm model;

[0014] Select the answer of the policy algorithm from the candidate options of the standard structured data of the selected questions through pre-configured rule policies.

[0015] As an alternative embodiment of the present invention, in the automatic problem-solving method for selected questions, the process of selecting the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected questions through the deep learning classification algorithm model includes:

[0016] Use the encoder network structure model based on transformer as the basic model structure;

[0017] Perform pre-training on the basic model structure based on a large amount of selected question bank data to obtain the deep learning classification algorithm model;

[0018] Perform binary classification model training on the standard structured data of the selected questions based on the deep learning classification algorithm model, and select the candidate option with the highest score as the answer of the deep learning classification algorithm.

[0019] As an alternative embodiment of the present invention, in the automatic problem-solving method for selected questions, the process of selecting the answer of the policy algorithm from the candidate options of the standard structured data of the selected questions through pre-configured rule policies includes:

[0020] Pre-configure a series of rule policies to be executed sequentially;

[0021] Sequentially execute the rule policies according to the configured order based on the standard structured data of the selected questions. If a certain rule policy gives a result, return it, stop the execution of subsequent rule policies, and use the returned result as the answer of the policy algorithm; if all rule policies are executed in order and no result is given, return "no answer of the policy algorithm".

[0022] As an alternative embodiment of the present invention, in the automatic problem-solving method for selected questions, the process of selecting the answer of the policy algorithm from the candidate options of the standard structured data of the selected questions through pre-configured rule policies includes:

[0023] Match according to the standard structured data for selecting test questions in the test question resource library for selection;

[0024] If the match is successful, obtain the answer from the information of the test question resources successfully matched in the test question resource library for selection, and use the obtained answer as the answer of the strategy algorithm; if the match fails, return the information of "automatic problem-solving failed".

[0025] As an optional implementation manner of the present invention, in the automatic problem-solving method for selecting test questions, the generating the final automatic problem-solving answer by integrating the answers of the deep learning classification algorithm and the answers of the strategy algorithm includes:

[0026] Judge whether the answer of the deep learning classification algorithm is consistent with the answer of the strategy algorithm. If the judgment result is yes, use the answer of the deep learning classification algorithm or the answer of the strategy algorithm as the final automatic problem-solving answer; if the judgment result is no, further judge whether the strategy algorithm obtains an answer. If the judgment result is yes, use the answer of the strategy algorithm as the final automatic problem-solving answer. If the judgment result is no, use the answer of the deep learning classification algorithm as the final automatic problem-solving answer.

[0027] As an optional implementation manner of the present invention, in the automatic problem-solving method for selecting test questions, the generating the final automatic problem-solving answer by integrating the answers of the deep learning classification algorithm and the answers of the strategy algorithm includes:

[0028] Preset the reference score for the answer selected by the deep learning classification algorithm;

[0029] When the score of the answer selected by the deep learning classification algorithm is higher than or equal to the reference score, directly use the answer of the deep learning classification algorithm as the final automatic problem-solving answer;

[0030] When the score of the answer selected by the deep learning classification algorithm is lower than the reference score, further judge whether the answer of the deep learning classification algorithm is consistent with the answer of the strategy algorithm. If the judgment result is yes, use the answer of the deep learning classification algorithm or the answer of the strategy algorithm as the final automatic problem-solving answer; if the judgment result is no, return the information of "automatic problem-solving failed".

[0031] As an optional implementation manner of the present invention, in the automatic problem-solving method for selecting test questions, the obtaining the text structured data of the test questions for selection includes:

[0032] Receive the input data of the request for automatic problem-solving of the test questions for selection, and the input data supports picture data and text data;

[0033] Judge whether the data is picture data;

[0034] If the judgment result is yes, the picture data is subjected to OCR recognition algorithm to generate fine-grained text structured data. If the judgment result is no, the input text data is directly used as the text structured data for the selected questions.

[0035] As an alternative implementation of the present invention, in the automatic problem-solving method for selected questions, the corresponding data processing for the text structured data to generate unified standard structured data includes:

[0036] For the text structured data, when it is judged that it contains multiple selected questions, the question splitting process is performed to generate text structured data for multiple single selected questions.

[0037] As an alternative implementation of the present invention, in the automatic problem-solving method for selected questions, the corresponding data processing for the text structured data to generate unified standard structured data includes:

[0038] Text standardization processing is performed on the text structured data to generate unified standard structured data. The text standardization processing performs annotation processing on the question attribute information and question content information in the text structured data of the selected questions;

[0039] Among them, the question attribute information includes question ID, and / or topic ID, and / or grade ID, and the question content information includes question text content and question option content.

[0040] The present invention also provides an automatic problem-solving device for selected questions, including:

[0041] An acquisition module that receives the input data of the request for automatic problem-solving of selected questions and acquires the text structured data of the selected questions in the input data;

[0042] An algorithm processing module that obtains the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data;

[0043] A comprehensive processing module that comprehensively generates the final automatic problem-solving answer based on the answers of the deep learning classification algorithm and the answers of the strategy algorithm.

[0044] The present invention also provides an electronic device, including a processor and a memory, where the memory is used to store computer-executable programs, and is characterized in that:

[0045] When the computer program is executed by the processor, the processor executes the automatic problem-solving method for selected questions.

[0046] The present invention also provides a computer-readable storage medium storing a computer-executable program, characterized in that when the computer-executable program is executed, the automatic problem-solving method for the selected questions is implemented.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] For the automatic problem-solving method of the selected questions in the present invention, regardless of the method used to obtain the selected questions, such as taking a photo of the question or text input, data processing needs to be performed to generate unified standard structured data. With such unified standard structured data, automatic answering can be achieved through the automatic problem-solving algorithm. Thus:

[0049] On the one hand, the application scenarios are expanded, and the user's learning cost for using is reduced. The user no longer needs to be limited to a specific method for inputting selected questions and can choose to take a photo of the question or enter text according to their own situation.

[0050] On the other hand, the availability of the automatic problem-solving algorithm for the selected questions obtained by various methods is ensured, making the automatic problem-solving method for the selected questions in this embodiment more reliable and practical.

[0051] For the automatic problem-solving method of the selected questions in the present invention, the automatic answering algorithm for the selected questions adopts a combination of a deep learning classification algorithm and a rule-based strategy algorithm. The deep learning classification algorithm and the rule-based strategy algorithm independently solve the selected questions. The deep learning classification algorithm and the rule-based strategy algorithm adopt different automatic answering logics, respectively generating the answers of the deep learning classification algorithm and the answers of the strategy algorithm. By integrating the answers generated by different answering logics, the final automatic problem-solving answer is generated. The two answering algorithms confirm and reference each other, thereby ensuring that a highly accurate automatic problem-solving answer can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 An example of a question type of the selected questions in an embodiment of the present invention;

[0053] Figure 2 A flowchart of the automatic problem-solving method for the selected questions in an embodiment of the present invention;

[0054] Figure 3 A module diagram of the automatic problem-solving device for the selected questions in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.

[0056] Accordingly, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely represents some embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0057] It should be noted that, without conflict, the embodiments in the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0058] It should be noted that similar reference numerals and letters indicate similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0059] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper", "lower", etc. is based on the orientation or positional relationship shown in the figures, or the orientation or positional relationship in which the product of the invention is usually placed during use, or the orientation or positional relationship commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention. In addition, terms such as "first", "second", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0060] See Figure 2 As shown, an automatic problem-solving method for multiple-choice questions in this embodiment includes:

[0061] Obtain the text structured data of the multiple-choice questions;

[0062] Based on the text structured data, obtain the answers of the deep learning classification algorithm and the answers of the strategy algorithm;

[0063] Integrate the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic problem-solving answer.

[0064] For the automatic problem-solving method of multiple-choice questions in this embodiment, regardless of the method used to obtain the multiple-choice questions, such as taking a photo of the question or text input, data processing needs to be performed to generate unified standard structured data. In this way, the unified standard structured data can be automatically answered through the automatic problem-solving algorithm. Thus:

[0065] On the one hand, the application scenarios are expanded, and the user's learning cost of use is reduced. The user no longer needs to be limited to a certain specified method for inputting multiple-choice questions and can choose to take a photo of the question or text input according to their own situation.

[0066] On the other hand, the availability of the automatic problem-solving algorithm for the selected test questions obtained in various ways is ensured, making the automatic problem-solving method for the selected test questions in this embodiment more reliable and practical.

[0067] For the automatic problem-solving method of the selected test questions in this embodiment, the deep learning classification algorithm and the rule strategy algorithm are combined for the automatic answering algorithm of the selected test questions. The deep learning classification algorithm and the rule strategy algorithm independently answer the selected test questions. The deep learning classification algorithm and the rule strategy algorithm adopt different automatic answering logics, respectively generate the answers of the deep learning classification algorithm and the answers of the strategy algorithm, and synthesize the answers generated by different answering logics to generate the final automatic problem-solving answer. The two answering algorithms confirm and reference each other, so as to ensure that an automatic problem-solving answer with high accuracy can be obtained.

[0068] Furthermore, obtaining the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data in this embodiment includes:

[0069] Perform corresponding data processing on the text structured data to generate unified standard structured data;

[0070] Select the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected test questions through the deep learning classification algorithm model;

[0071] Select the answer of the strategy algorithm from the candidate options of the standard structured data of the selected test questions through the pre-configured rule strategy;

[0072] Synthesize the answer of the deep learning classification algorithm and the answer of the strategy algorithm to generate the final automatic problem-solving answer.

[0073] As an optional implementation manner of this embodiment, in the automatic problem-solving method of a selected test question in this embodiment, the process of selecting the answer of the deep learning classification algorithm from the candidate options in the standard structured data of the selected test questions through the deep learning classification algorithm model includes:

[0074] Use the encoder network structure model based on transformer as the basic model structure;

[0075] Pre-train the basic model structure based on the massive selected test question bank data in the education field to obtain the deep learning classification algorithm model;

[0076] Perform binary classification model training on the standard structured data of the selected test questions based on the deep learning classification algorithm model, and select the candidate option with the highest score as the answer of the deep learning classification algorithm.

[0077] Transformer is a brand-new encoder-decoder model based on the self-attention mechanism. Compared with similar models built by traditional recurrent neural networks, Transformer has multiple advantages: solving the problem of long-term dependence, being parallelizable, etc. Specifically, for example, BERT, ERNIE, etc. can be selected based on the Transformer encoder.

[0078] The full name of BERT is Bidirectional Encoder Representation from Transformers. BERT uses the Transformer Encoder block for connection because it is a typical bidirectional encoding model.

[0079] ERNIE: Enhanced Representation through Knowledge Integration is a further optimization based on the BERT model and has achieved state-of-the-art results in Chinese NLP tasks. It mainly makes improvements in the masking mechanism. Its mask is not the basic word piece mask, but additional external knowledge is added during the pre-training stage, consisting of three levels of masking, namely basic-level masking (wordpiece) + phrase level masking (WWM style) + entity level masking. On this basis, Chinese ERNIE also uses various heterogeneous datasets.

[0080] This embodiment uses the Transformer encoder network structure model as the basic model structure. The Transformer encoder network structure model has powerful language representation and feature extraction capabilities, and can understand the questions and extract features for the selected test questions. In this embodiment, after selecting the Transformer encoder network structure model as the basic model structure, pre-training is performed on the basic model structure based on the massive data of the selected test question bank in the education field to obtain a deep learning classification algorithm model; in this way, the deep learning classification algorithm model can automatically answer the selected test questions in the education field.

[0081] Furthermore, in this embodiment, the standard structured data for selecting questions is used to train a binary classification model based on the deep learning classification algorithm model, and the candidate option with the highest score is selected as the answer of the deep learning classification algorithm. Each candidate option in the standard structured data for selecting questions is input into the deep learning classification algorithm model to calculate the score, and the candidate option with the highest score is selected through the binary classification model algorithm as the automatic answer to the selected question.

[0082] As an alternative implementation of this embodiment, an automatic question-solving method for selected questions in this embodiment selects the answer of the policy algorithm from the candidate options of the standard structured data for selected questions through pre-configured rule policies, including:

[0083] Pre-configure a series of rule policies to be executed sequentially;

[0084] According to the standard structured data of the selected questions, the rule policies are executed sequentially according to the configured order. If a certain rule policy gives a result, then return, stop the execution of the subsequent rule policies, and use the returned result as the answer of the policy algorithm; if all the rule policies are executed in order and no result is given, then return "no answer of the policy algorithm".

[0085] Among them, the rule policies include part-of-speech rules, type rules, pronunciation rules, and can be extended. Taking the first part-of-speech rule as an example: A. homework, B. tomorrow, C. today, where A. homework is a noun, B. tomorrow and C. today are adverbs, then select A. Similar for other rules. As long as a certain rule gives a result in order, then return, and there is no need to continue with the subsequent rules.

[0086] As an alternative implementation of this embodiment, an automatic question-solving method for selected questions in this embodiment selects the answer of the policy algorithm from the candidate options of the standard structured data for selected questions through pre-configured rule policies, including:

[0087] Match according to the standard structured data of the selected questions in the selected question resource library;

[0088] If the match is successful, obtain the answer from the question resource information that matches successfully in the selected question resource library, and use the obtained answer as the answer of the policy algorithm; if the match fails, then return the information "automatic question-solving failed".

[0089] The rule-based strategy algorithm in this embodiment uses a different solution logic from the deep learning classification algorithm to automatically solve test questions. The deep learning classification algorithm is trained based on a neural network to achieve automatic solution of multiple-choice test questions. The rule-based strategy algorithm, on the other hand, performs matching based on a multiple-choice test question resource library to achieve automatic solution. The two algorithms achieve automatic solution of multiple-choice test questions through different methods, and then the final automatic test question solution answer is obtained by integrating the solution results of the two algorithms. On the one hand, the two solution algorithms verify each other, which can increase the reliability of automatic test question solution and ensure the accuracy of the answer to the automatic test question solution. On the other hand, the two solution algorithms complement each other, which can increase the success rate of test question solution, ensure obtaining the automatic solution answer of multiple-choice test questions, and ensure the usability of the automatic test question solving method for multiple-choice test questions in this embodiment.

[0090] Specifically, the matching of the standard structured data of the multiple-choice test questions in the multiple-choice test question resource library in this embodiment includes:

[0091] Obtain the test question attribute information in the standard structured data;

[0092] Perform matching in the multiple-choice test question resource library according to the test question attribute information.

[0093] Specifically, the test question attribute information includes year attribute information, and / or subject attribute information, and / or grade attribute information, and / or semester attribute information, and / or region attribute information, and / or test paper type attribute information;

[0094] Obtain the year matching information, and / or subject matching information, and / or subject matching information, and / or grade matching information, and / or semester matching information, and / or region matching information, and / or test paper type matching information in the test question attribute information;

[0095] Match each of the matching information with the corresponding attribute information in the multiple-choice test question resource library through a string matching algorithm, and output the matching results respectively.

[0096] In addition, it is also possible to perform coincidence matching of the test question content information in the standard structured data of the multiple-choice test questions in the multiple-choice test question resource library. When the coincidence degree is greater than the preset value, it is determined that the matching is successful; otherwise, it is determined that the matching fails.

[0097] As an optional implementation manner of this embodiment, in an automatic test question solving method for multiple-choice test questions in this embodiment, the generation of the final automatic test question solution answer by integrating the answers of the deep learning classification algorithm and the strategy algorithm includes:

[0098] Determine whether the answer of the deep learning classification algorithm is the same as the answer of the policy algorithm. If the determination result is yes, use the answer of the deep learning classification algorithm or the answer of the policy algorithm as the final automatic problem-solving answer. If the determination result is no, further determine whether the rule policy algorithm selects the answer of the policy algorithm. If the determination result is yes, use the answer of the policy algorithm as the final automatic problem-solving answer. If the determination result is no, use the answer of the deep learning classification algorithm as the final automatic problem-solving answer.

[0099] In this embodiment, by determining whether the answers of the two question-solving algorithms are the same, the two algorithms implement the automatic solution of multiple-choice questions in different ways, and then obtain the final automatic question-solving answer by integrating the solution results of the two algorithms. On the one hand, the two solution algorithms verify each other, which can increase the reliability of the automatic question-solving and ensure the accuracy of the answer to the automatic question-solving. On the other hand, the two solution algorithms complement each other, which can increase the success rate of question-solving and ensure that the automatic solution answer of multiple-choice questions is obtained, thereby ensuring the usability of the automatic problem-solving method for multiple-choice questions in this embodiment.

[0100] As an optional implementation manner of this embodiment, for an automatic problem-solving method for multiple-choice questions in this embodiment, the generating the final automatic problem-solving answer by integrating the answer of the deep learning classification algorithm and the answer of the policy algorithm includes:

[0101] Preset a reference score for the answer selected by the deep learning classification algorithm;

[0102] When the score of the answer selected by the deep learning classification algorithm is higher than or equal to the reference score, directly use the answer of the deep learning classification algorithm as the final automatic problem-solving answer;

[0103] When the score of the answer selected by the deep learning classification algorithm is lower than the reference score, further determine whether the answer of the deep learning classification algorithm is the same as the answer of the policy algorithm. If the determination result is yes, use the answer of the deep learning classification algorithm or the answer of the policy algorithm as the final automatic problem-solving answer. If the determination result is no, return the information of "automatic problem-solving failed".

[0104] As an optional implementation manner of this embodiment, for an automatic problem-solving method for multiple-choice questions in this embodiment, the generating the final automatic problem-solving answer by integrating the answer of the deep learning classification algorithm and the answer of the policy algorithm includes:

[0105] Due to the high recall rate of the deep learning classification algorithm and the high recall accuracy of the policy algorithm, the deep learning classification algorithm and the policy algorithm are fused in the way that the policy algorithm takes precedence. That is, if the policy algorithm gives a problem-solving result, it shall prevail; if the policy algorithm has no result, the answer result of the deep learning classification algorithm will be used instead.

[0106] In the process of generating the final automatic problem-solving answer by integrating the answers of the deep learning classification algorithm and the policy algorithm in this embodiment, first, based on the deep learning classification algorithm, when the score of the answer selected by the deep learning classification algorithm is greater than the reference score, it is considered that the answer obtained by the deep learning classification algorithm has a relatively high credibility and can be directly used as the automatic answer for selecting the test question. When the score of the answer selected by the deep learning classification algorithm is less than the reference score and the credibility of the answer obtained by the deep learning classification algorithm is not so high, the answer of the rule-based policy algorithm is further combined to judge whether they are the same. If they are the same, it is considered that the answers obtained by the two algorithms are relatively reliable and used as the automatic answer for selecting the test question. If they are different, it is considered that there is a relatively low credibility for the automatic answer to the currently selected test question, and the automatic answers of the two algorithms cannot be used as the automatic answer to the current test question, and the information of "automatic problem-solving failed" is reported.

[0107] As an optional implementation manner of this embodiment, in an automatic problem-solving method for selecting test questions in this embodiment, the obtaining of the text structured data of the selected test question includes:

[0108] Receiving the input data of the request for automatic problem-solving of the selected test question, and the input data supports picture data and text data;

[0109] Judging whether the data is picture data;

[0110] If the judgment result is yes, the picture data is processed through the ocr recognition algorithm to generate fine-grained text structured data. If the judgment result is no, the input text data is directly used as the text structured data of the selected test question.

[0111] The input data that can be received by the test question automatic answering method of this embodiment includes test question pictures and test question texts. The test question pictures can be test question photos taken as shown in Figure 1 and the examples of the test question text data can be:

[0112] i.()4.It was nice__your email.A.got B.get C.to get

[0113] ii.()5.__do we have Chinese this morning?A.Who B.How C.What D.When。

[0114] Furthermore, in the automatic problem-solving method for selected questions described in this embodiment, the corresponding data processing for the text structured data to generate unified standard structured data includes:

[0115] For the text structured data, when it is determined that it contains multiple selected questions, it is split into multiple text structured data of single selected questions.

[0116] Specifically, in the automatic problem-solving method for selected questions in this embodiment, the corresponding data processing for the text structured data to generate unified standard structured data includes:

[0117] Perform text standardization processing on the text structured data to generate unified standard structured data. The text standardization processing performs annotation processing on the question attribute information and question content information in the text structured data of the selected questions;

[0118] Among them, the question attribute information includes question ID, and / or topic ID, and / or grade ID, and the question content information includes question text content and question option content.

[0119] A specific example of the corresponding data processing for the text structured data in this embodiment to generate unified standard structured data is:

[0120] Input: 5()Teddy is________to our city, he just come here yesterday..A.old B.new C.young D short; After data processing, unified standard structured data is generated, and the output is:

[0121]

[0122]

[0123] See Figure 3 As shown, this embodiment also provides an automatic problem-solving device for selected questions, including:

[0124] An acquisition module that receives the input data of the automatic problem-solving request for the selected questions and acquires the text structured data of the selected questions in the input data;

[0125] An algorithm processing module that obtains the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data;

[0126] The comprehensive processing module combines the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic problem-solving answer.

[0127] Further, the automatic problem-solving device for selected questions in this embodiment includes:

[0128] The data structuring processing module performs corresponding data processing on the text structured data to generate unified standard structured data;

[0129] The algorithm processing module includes a deep learning classification module and a strategy module;

[0130] The deep learning classification module selects the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected questions through the deep learning classification algorithm model;

[0131] The strategy module selects the answer of the strategy algorithm from the candidate options of the standard structured data of the selected questions through the pre-configured rule strategy.

[0132] For the automatic problem-solving device for selected questions in this embodiment, regardless of the method used to obtain the selected questions, such as taking a photo of the question or text input, they can be obtained by the acquisition module, and the obtained selected questions need to be processed by the data structuring processing module to generate unified standard structured data. In this way, the unified standard structured data can be automatically answered through the algorithm model module, so that:

[0133] On the one hand, the application scenario is expanded, and the user's usage and learning costs are reduced. The user no longer needs to be limited to a certain specified method for entering selected questions and can choose to take a photo of the question or enter text according to their own situation.

[0134] On the other hand, the availability of the automatic problem-solving algorithm for selected questions obtained by various methods is ensured, making the automatic problem-solving device for selected questions in this embodiment more reliable and practical.

[0135] For the automatic problem-solving device for selected questions in this embodiment, the algorithm model module includes a deep learning classification module and a strategy module. The automatic problem-solving algorithm for selected questions uses a combination of the deep learning classification algorithm and the rule strategy algorithm. The deep learning classification algorithm and the rule strategy algorithm independently select questions to answer. The deep learning classification algorithm and the rule strategy algorithm use different automatic answering logics to generate the answers of the deep learning classification algorithm and the strategy algorithm respectively, and combine the answers generated by different answering logics to generate the final automatic problem-solving answer. The two answering algorithms confirm and reference each other, so as to ensure that a highly accurate automatic problem-solving answer can be obtained.

[0136] As an alternative implementation of this embodiment, the deep learning classification module in this embodiment selects the answer of the deep learning classification algorithm from the candidate options in the standard structured data of the selected questions, including:

[0137] The deep learning classification algorithm model uses the encoder network structure model based on the transformer as the basic model structure;

[0138] Pre-train the basic model structure based on the massive selection question bank data in the education field to obtain the deep learning classification algorithm model;

[0139] Perform binary classification model training on the standard structured data of the selected questions based on the deep learning classification algorithm model, and select the candidate option with the highest score as the answer of the deep learning classification algorithm.

[0140] In this embodiment, the encoder network structure model based on the transformer is used as the basic model structure. The encoder network structure model of the transformer has powerful language representation ability and feature extraction ability, and can perform question understanding and feature extraction for the selected questions. In this embodiment, when selecting the encoder network structure model based on the transformer as the basic model structure, and then pre-training the basic model structure based on the massive selection question bank data in the education field to obtain the deep learning classification algorithm model; in this way, the deep learning classification algorithm model can realize the automatic answering of the selection questions in the education field.

[0141] Furthermore, in this embodiment, binary classification model training is performed on the standard structured data of the selected questions based on the deep learning classification algorithm model, and the candidate option with the highest score is selected as the answer of the deep learning classification algorithm. Input each candidate option in the standard structured data of the selected questions into the deep learning classification algorithm model to calculate the score, and select the candidate option with the highest score through the binary classification model algorithm as the automatic answering answer of the selected question.

[0142] As an alternative implementation of this embodiment, the strategy module in this embodiment selects the answer of the strategy algorithm from the candidate options in the standard structured data of the selected questions based on the selected question resource library, including:

[0143] Match according to the standard structured data of the selected questions in the selected question resource library;

[0144] If the match is successful, obtain the answer from the question resource information successfully matched in the selected question resource library, and use the obtained answer as the answer of the strategy algorithm; if the match fails, return the information "Automatic question-solving failed".

[0145] The strategy module in this embodiment uses a different solution logic from the deep learning classification module to automatically solve test questions. The deep learning classification module is trained based on a neural network to achieve automatic solution of multiple-choice test questions. The strategy module, on the other hand, performs matching based on a multiple-choice test question resource library to perform automatic solution. The two algorithms achieve automatic solution of multiple-choice test questions through different methods, and then the solution results of the two modules are combined to obtain the final automatic test question solution answer. On the one hand, the two solution modules verify each other, which can increase the reliability of the automatic test question solution and ensure the accuracy of the answer to the automatic test question solution; on the other hand, the two solution modules complement each other, which can increase the success rate of the test question solution and ensure obtaining the automatic solution answer to the multiple-choice test question, guaranteeing the usability of the automatic test question solving method for multiple-choice test questions in this embodiment.

[0146] As an optional implementation manner of this embodiment, the comprehensive processing module of this embodiment combines the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic test question solution answer, including:

[0147] Judge whether the answer of the deep learning classification algorithm is the same as the answer of the strategy algorithm. If the judgment result is yes, use the same answer of the deep learning classification algorithm and the answer of the strategy algorithm as the final automatic test question solution answer. If the judgment result is no, further judge whether the rule strategy algorithm selects the answer of the strategy algorithm. If the judgment result is yes, use the answer of the strategy algorithm as the final automatic test question solution answer. If the judgment result is no, use the answer of the deep learning classification algorithm as the final automatic test question solution answer.

[0148] In this embodiment, by judging whether the answers of the two test question solution algorithms are the same, the two algorithms achieve automatic solution of multiple-choice test questions through different methods, and then the solution results of the two algorithms are combined to obtain the final automatic test question solution answer. On the one hand, the two solution algorithms verify each other, which can increase the reliability of the automatic test question solution and ensure the accuracy of the answer to the automatic test question solution; on the other hand, the two solution algorithms complement each other, which can increase the success rate of the test question solution and ensure obtaining the automatic solution answer to the multiple-choice test question, guaranteeing the usability of the automatic test question solving method for multiple-choice test questions in this embodiment.

[0149] As an optional implementation manner of this embodiment, the comprehensive processing module of this embodiment combines the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic test question solution answer, including:

[0150] Preset the reference score for the answer selected by the deep learning classification algorithm;

[0151] When the score of the answer selected by the deep learning classification algorithm is higher than or equal to the reference score, the answer of the deep learning classification algorithm is directly used as the final automatic problem-solving answer;

[0152] When the score of the answer selected by the deep learning classification algorithm is lower than the reference score, it is further determined whether the answer of the deep learning classification algorithm is consistent with the answer of the policy algorithm. If the determination result is yes, the answer of the deep learning classification algorithm or the answer of the policy algorithm is used as the final automatic problem-solving answer. If the determination result is no, an "automatic problem-solving failed" message is returned.

[0153] In the process of the comprehensive module of this embodiment synthesizing the answers of the deep learning classification algorithm and the policy algorithm to generate the final automatic problem-solving answer, first, based on the deep learning classification algorithm, when the score of the answer selected by the deep learning classification algorithm is greater than the reference score, it is considered that the answer obtained by the deep learning classification algorithm has a higher credibility and can be directly used as the automatic answer for selecting the test question. When the score of the answer selected by the deep learning classification algorithm is less than the reference score and the credibility of the answer obtained by the deep learning classification algorithm is not so high, then further combine the answer of the rule-based policy algorithm to determine whether they are the same. If they are the same, it is considered that the answers obtained by the two algorithms are relatively reliable and used as the automatic answer for selecting the test question. If they are not the same, it is considered that the automatic answer for the currently selected test question has a relatively low credibility, and the automatic answers of the two algorithms cannot be used as the automatic answer for the current test question, and an "automatic problem-solving failed" message is returned.

[0154] As an optional implementation manner of this embodiment, the acquisition module of this embodiment receives the input data of the request for automatically solving the selected test question, and the text structured data of the selected test question obtained from the input data includes:

[0155] Receive the input data of the request for automatically solving the selected test question, and the input data supports picture data and text data;

[0156] Determine whether the data is picture data;

[0157] If the determination result is yes, the picture data is recognized by the ocr recognition module to generate fine-grained text structured data. If the determination result is no, the input text data is directly used as the text structured data of the selected test question.

[0158] Furthermore, the data structured processing module of this embodiment performs corresponding data processing on the text structured data to generate unified standard structured data, including:

[0159] For the text structured data, when it is determined that it contains multiple-choice questions, it is split into multiple single-choice question text structured data.

[0160] Specifically, the data structuring and processing module of this embodiment performs corresponding data processing on the text structured data to generate unified standard structured data, including:

[0161] Perform text standardization processing on the text structured data to generate unified standard structured data. The text standardization processing performs annotation processing on the question attribute information and question content information in the text structured data of the multiple-choice questions;

[0162] Among them, the question attribute information includes question ID, and / or topic ID, and / or grade ID, and the question content information includes question text content and question option content.

[0163] This embodiment also provides a computer-readable storage medium storing a computer-executable program, which when executed, implements the automatic problem-solving method for the multiple-choice questions.

[0164] The computer-readable storage medium in this embodiment may include a data signal propagated in a baseband or as part of a carrier wave, which carries readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable storage medium may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination of the above.

[0165] This embodiment also provides an electronic device, including a processor and a memory. The memory is used to store a computer-executable program, and when the computer program is executed by the processor, the processor executes the automatic problem-solving method for the multiple-choice questions.

[0166] The electronic device is presented in the form of a general-purpose computing device. The processor may be one or multiple and work cooperatively. The present invention does not exclude distributed processing, that is, the processors may be dispersed in different physical devices. The electronic device of the present invention is not limited to a single entity, but may also be the sum of multiple physical devices.

[0167] The memory stores a computer-executable program, usually machine-readable code. The computer-readable program can be executed by the processor so that the electronic device can execute the method of the present invention, or at least some steps of the method.

[0168] The memory includes volatile memory, such as random access memory units (RAM) and / or cache memory units, and may also include non-volatile memory, such as read-only memory units (ROM).

[0169] It should be understood that the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as display screens, and some electronic devices also include human-computer interaction elements, such as buttons, keyboards, etc. As long as the electronic device can execute the computer-readable program in the memory to implement at least part of the steps of the method of the present invention, it can be considered as the electronic device covered by the present invention.

[0170] From the above description of the embodiments, those skilled in the art can easily understand that the present invention can be implemented by hardware capable of executing a specific computer program, such as the system of the present invention, and the electronic processing units, servers, clients, mobile phones, control units, processors, etc. included in the system. The present invention can also be implemented by computer software for executing the method of the present invention, such as control software executed by a microprocessor, an electronic control unit, a client, a server, etc. However, it should be noted that the computer software for executing the method of the present invention is not limited to being executed in one or a specific number of hardware entities, and it can also be implemented in a distributed manner by unspecified specific hardware. For computer software, the software product can be stored in a computer-readable storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), or can be distributed and stored on a network, as long as it can enable the electronic device to execute the method according to the present invention.

[0171] The above embodiments are only used to illustrate the present invention and do not limit the technical solutions described in the present invention. Although the present specification has described the present invention in detail with reference to the above respective embodiments, the present invention is not limited to the above specific embodiments. Therefore, any modification or equivalent replacement of the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the invention are covered by the scope of the claims of the present invention.

Claims

1. An automatic problem-solving method for multiple-choice questions, characterized in that, it includes: Obtain the text structured data of the multiple-choice questions; Based on the text structured data, obtain the answers of the deep learning classification algorithm and the answers of the strategy algorithm; Integrate the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic problem-solving answer; The obtaining the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data includes: Perform corresponding data processing on the text structured data to generate unified standard structured data; Select the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the multiple-choice questions through the deep learning classification algorithm model; Select the answer of the strategy algorithm from the candidate options of the standard structured data of the multiple-choice questions through the pre-configured rule strategy; The selecting the answer of the deep learning classification algorithm from the candidate options in the standard structured data of the multiple-choice questions through the deep learning classification algorithm model includes: Use the encoder network structure model based on transformer as the basic model structure; Perform pre-training on the basic model structure based on a large amount of multiple-choice question bank data to obtain the deep learning classification algorithm model; Perform binary classification model training on the standard structured data of the multiple-choice questions based on the deep learning classification algorithm model, and select the candidate option with the highest score as the answer of the deep learning classification algorithm; The selecting the answer of the strategy algorithm from the candidate options of the standard structured data of the multiple-choice questions through the pre-configured rule strategy includes: Pre-configure a series of rule strategies to be executed sequentially; According to the standard structured data of the multiple-choice questions, sequentially execute the rule strategies in the configured order. If a certain rule strategy gives a result, return it, stop the execution of the subsequent rule strategies, and use the returned result as the answer of the strategy algorithm; if no result is given after sequentially executing all the rule strategies, return the answer without a strategy algorithm; The integrating the answers of the deep learning classification algorithm and the answers of the strategy algorithm to generate the final automatic problem-solving answer includes: Preset the reference score for the answer selected by the deep learning classification algorithm; When the score of the answer selected by the deep learning classification algorithm is higher than or equal to the reference score, directly use the answer of the deep learning classification algorithm as the final automatic problem-solving answer; When the score of the answer selected by the deep learning classification algorithm is lower than the reference score, further determine whether the answer of the deep learning classification algorithm is consistent with the answer of the strategy algorithm. If the judgment result is yes, use the answer of the deep learning classification algorithm or the answer of the strategy algorithm as the final automatic problem-solving answer; if the judgment result is no, return the automatic problem-solving failure information.

2. The automatic problem-solving method for multiple-choice questions according to claim 1, characterized in that, The selecting the answer of the strategy algorithm from the candidate options of the standard structured data of the multiple-choice questions through the pre-configured rule strategy is also implemented by the following method: Match according to the standard structured data of the multiple-choice questions in the multiple-choice question resource library; If the matching is successful, obtain the answer from the information of the question resource that matches successfully in the selected question resource library, and use the obtained answer as the answer of the strategy algorithm; if the matching fails, return the information indicating that the automatic problem-solving fails.

3. The automatic problem-solving method for selected questions according to claim 1, characterized in that the generation of the final automatic problem-solving answer by integrating the answer of the deep learning classification algorithm and the answer of the strategy algorithm is also achieved by the following method: Judge whether the answer of the deep learning classification algorithm is consistent with the answer of the strategy algorithm. If the judgment result is yes, use the answer of the deep learning classification algorithm or the answer of the strategy algorithm as the final automatic problem-solving answer; if the judgment result is no, further judge whether the strategy algorithm gets an answer. If the judgment result is yes, use the answer of the strategy algorithm as the final automatic problem-solving answer. If the judgment result is no, use the answer of the deep learning classification algorithm as the final automatic problem-solving answer.

4. The automatic problem-solving method for selected questions according to claim 1, characterized in that the obtaining of the text structured data of the selected questions includes: Receiving the input data of the request for automatic problem-solving of selected questions, where the input data supports picture data and text data; Judging whether the data is picture data; If the judgment result is yes, perform ocr recognition algorithm on the picture data to generate fine-grained text structured data; if the judgment result is no, directly use the input text data as the text structured data of the selected questions.

5. The automatic problem-solving method for selected questions according to claim 1, characterized in that the corresponding data processing for the text structured data to generate unified standard structured data includes: For the text structured data, when it is judged that it contains multiple selected questions, perform question splitting processing to generate text structured data of multiple single selected questions.

6. The automatic problem-solving method for selected questions according to claim 5, characterized in that the corresponding data processing for the text structured data to generate unified standard structured data includes: Perform text standardization processing on the text structured data to generate unified standard structured data. The text standardization processing performs standardization processing on the question attribute information and question content information in the text structured data of the selected questions; wherein, the question attribute information includes question ID, and / or topic ID, and / or grade ID, and the question content information includes question text content and question option content.

7. An automatic problem-solving device for selected questions, characterized in that it includes: An acquisition module that receives the input data of the request for automatic problem-solving of selected questions and acquires the text structured data of the selected questions in the input data; An algorithm processing module that obtains the answer of the deep learning classification algorithm and the answer of the strategy algorithm based on the text structured data; A comprehensive processing module that integrates the answer of the deep learning classification algorithm and the answer of the strategy algorithm to generate the final automatic problem-solving answer; Obtaining the answers of the deep learning classification algorithm and the answers of the strategy algorithm based on the text structured data includes: Performing corresponding data processing on the text structured data to generate unified standard structured data; Selecting the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected questions through the deep learning classification algorithm model; Selecting the answer of the strategy algorithm from the candidate options of the standard structured data of the selected questions through the pre-configured rule strategy; The step of selecting the answer of the deep learning classification algorithm from the candidate options of the standard structured data of the selected questions through the deep learning classification algorithm model includes: Using the encoder network structure model based on transformer as the basic model structure; Performing pre-training on the basic model structure based on a large amount of selected question bank data to obtain the deep learning classification algorithm model; Performing binary classification model training on the standard structured data of the selected questions based on the deep learning classification algorithm model, and selecting the candidate option with the highest score as the answer of the deep learning classification algorithm; The step of selecting the answer of the strategy algorithm from the candidate options of the standard structured data of the selected questions through the pre-configured rule strategy includes: Pre-configuring a series of rule strategies to be executed sequentially; Sequentially executing the rule strategies according to the configured order based on the standard structured data of the selected questions. If a certain rule strategy gives a result, return it, stop the execution of the subsequent rule strategies, and use the returned result as the answer of the strategy algorithm; if no result is given after sequentially executing all the rule strategies, return the answer of no strategy algorithm; Generating the final automatic problem-solving answer by integrating the answer of the deep learning classification algorithm and the answer of the strategy algorithm includes: Presetting a reference score for the answer selected by the deep learning classification algorithm; When the score of the answer selected by the deep learning classification algorithm is higher than or equal to the reference score, directly use the answer of the deep learning classification algorithm as the final automatic problem-solving answer; When the score of the answer selected by the deep learning classification algorithm is lower than the reference score, further determine whether the answer of the deep learning classification algorithm is consistent with the answer of the strategy algorithm. If the judgment result is yes, use the answer of the deep learning classification algorithm or the answer of the strategy algorithm as the final automatic problem-solving answer; if the judgment result is no, return the automatic problem-solving failure information.

8. An electronic device, including a processor and a memory, the memory is used to store computer-executable programs, characterized in that: When the computer program is executed by the processor, the processor executes the automatic problem-solving method for selected questions according to any one of claims 1-6.

9. A computer-readable storage medium, storing a computer-executable program, characterized in that, When the computer-executable program is executed, the automatic problem-solving method for selected questions according to any one of claims 1-6 is implemented.

Images