Test question auxiliary learning explanation method and device, related equipment and computer program product

Abstract

The invention discloses a test question auxiliary learning explanation method and device, related equipment and a computer program product, and relates to the technical field of artificial intelligence. According to the method, for the target test question to be tutored, the auxiliary learning background information required by auxiliary learning explanation of the target test question is firstly acquired, such as a reference answer of the target test question, subject knowledge required by explanation, explanation step planning and the like; and calling the capability of the auxiliary learning large model, and generating auxiliary learning explanation content of the target test question according to the form of the semantic unit block in combination with the auxiliary learning background information and the configured field meaning of the semantic unit block. The semantic unit block comprises text draft fields, and the text draft fields are contents to be broadcasted in the explanation process. The text manuscript in the auxiliary learning explanation content is output in a voice broadcast mode, the oral content of the teacher in the class process is simulated, the students are guided to answer the test questions step by step, and the tutoring effect on the students is improved. According to the invention, students can be better guided to understand the test questions.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and more specifically, to a test question auxiliary learning explanation method, device, related equipment and computer program product. Background Art

[0002] When students are doing after-class exercises, there are mainly two types of user scenarios. One is that students are doing after-class exercises and need answers and solutions to questions they don’t know how to answer. The other is that parents are tutoring students in after-class exercises, but due to limited time, energy and teaching ability, it is difficult to provide targeted explanations.

[0003] Some homework assistance systems currently on the market only obtain reference answers to test questions and output them, lack supplementary explanation content, and have little effect in tutoring students. Summary of the invention

[0004] In view of the above problems, this application is proposed to provide a method, device, related equipment and computer program product for assisting students in learning and explaining test questions, so as to output assisting students in learning and explaining the target test questions to be tutored, thereby improving the tutoring effect on students. The specific plan is as follows:

[0005] In the first aspect, a method for explaining test questions is provided, including:

[0006] Obtain the background information needed to provide supplementary explanations for target test questions;

[0007] Calling the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block, wherein the semantic unit block at least includes a text script field, and the text script field is the content of the explanation process;

[0008] The text of the text field in the auxiliary explanation content is voice broadcasted.

[0009] In a possible design, in another implementation of the first aspect of the embodiments of the present application, the supplementary learning background information includes one or more of the following combinations:

[0010] Reference answers to the target test questions, subject knowledge required for explanation, and explanation step planning information.

[0011] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of obtaining the reference answer to the target test question includes:

[0012] The big model is called to instruct the big model to generate a reference answer to the target test question.

[0013] In a possible design, in another implementation of the first aspect of the embodiment of the present application, before calling the large model to generate a reference answer to the target test question, the method further includes:

[0014] Retrieving similar question data that meets similarity requirements with the target test question;

[0015] The process of calling the large model to generate a reference answer to the target test question includes:

[0016] The big model is called to instruct the big model to refer to the similar question data and generate a reference answer to the target test question.

[0017] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of acquiring the subject knowledge required for explaining the target test question includes:

[0018] Calling the configured retrieval condition generation model to generate the subject knowledge retrieval condition corresponding to the target test question, wherein the retrieval condition generation model is trained using sample test questions annotated with subject knowledge retrieval condition labels;

[0019] The subject knowledge search conditions are used to search for target subject knowledge in a subject knowledge base as the subject knowledge required for the target test question explanation process.

[0020] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the retrieval condition generation model is a retrieval condition generation large model obtained by training an initial large model using sample test questions annotated with subject knowledge retrieval condition tags;

[0021] The process of calling the retrieval condition generation macro model to generate the subject knowledge retrieval condition corresponding to the target test question includes:

[0022] Calling the retrieval condition to generate a large model to instruct the large model to combine the configured intent list and parameter definition, select the target intent to which the target test question belongs and generate parameter information of the target test question under the target intent, wherein the intent list includes the examination intent of each knowledge point under the subject to which the target test question belongs;

[0023] The target intention and the corresponding parameter information constitute the subject knowledge retrieval condition corresponding to the target test question.

[0024] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of using the subject knowledge search condition to search for target subject knowledge in the subject knowledge base includes:

[0025] If the subject knowledge search condition meets the template rule, the heuristic rule search strategy is used to search for the target subject knowledge in the subject knowledge base;

[0026] If the subject knowledge retrieval condition does not meet the template rule, the query vector representation corresponding to the subject knowledge retrieval condition is determined, and based on the vector similarity, the target subject knowledge that meets the similarity requirement with the query vector representation is determined in the subject knowledge base.

[0027] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of acquiring the explanation step planning information of the target test question includes:

[0028] The big model is called to instruct the big model to generate explanation step planning information for the target test question.

[0029] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the semantic unit block further includes: a display content field, the display content field is the blackboard content on the upper screen;

[0030] Then the auxiliary learning explanation content of the target test question generated by the auxiliary learning big model is called to also include the blackboard content corresponding to the display content field;

[0031] The method further includes:

[0032] On the basis of the voice broadcast of the text script, the blackboard content corresponding to the display content field is displayed on the screen.

[0033] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the semantic unit block further includes: a test question marking content field, the test question marking content field is a test question segment that needs to be marked when explaining the current content;

[0034] Then the auxiliary learning explanation content of the target test question generated by the auxiliary learning big model is called to also include the test question fragment corresponding to the test question marking content field;

[0035] The method further includes:

[0036] On the basis of the voice broadcast of the text script, the test question segment corresponding to the test question marking content field in the target test question is marked and displayed.

[0037] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the test question marking content field also includes a marking form;

[0038] The process of marking and displaying the test question segment corresponding to the test question marking content field in the target test question includes:

[0039] The test question segment corresponding to the test question marking content field in the target test question is marked and displayed in a corresponding marking format.

[0040] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the method further includes:

[0041] Obtain user feedback on the output supplementary teaching content;

[0042] The auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field meaning of the configured semantic unit block and the feedback content, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0043] In a possible design, in another implementation of the first aspect of the embodiment of the present application, before the voice broadcast of the text of the text field in the auxiliary explanation content is performed, it also includes:

[0044] Evaluate the supplementary teaching content generated this time and generate modification suggestions;

[0045] When the evaluation result indicates that the supplementary teaching explanation content generated this time meets the requirements, the step of voice broadcasting the text script in the text script field in the supplementary teaching explanation content is executed.

[0046] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the method further includes:

[0047] When the evaluation result indicates that the auxiliary learning explanation content generated this time does not meet the requirements, the auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field definition of the configured semantic unit block, the auxiliary learning explanation content generated this time, the evaluation result and the modification suggestion, and regenerate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0048] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of evaluating the auxiliary learning explanation content generated this time and generating modification suggestions includes:

[0049] The big model is called to instruct the big model to evaluate the supplementary teaching content generated this time and generate modification suggestions.

[0050] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of evaluating the auxiliary learning explanation content generated this time is evaluated according to the configured evaluation dimension, and the evaluation dimension includes at least one of the following:

[0051] Whether the evaluation of students' answers is accurate, whether there are knowledge errors in the explanation, whether there are logical errors in the explanation, and whether the explanation can promote students' understanding of the solution ideas of the target test questions.

[0052] In a second aspect, a test question auxiliary learning explanation device is provided, comprising:

[0053] A supplementary learning background information acquisition unit is used to acquire supplementary learning background information required for supplementary learning explanation of the target test question, wherein the supplementary learning background information includes: a reference answer to the target test question, subject knowledge required for explanation, and explanation step planning;

[0054] The auxiliary learning explanation content generation unit is used to call the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block to generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block, wherein the semantic unit block at least includes a text manuscript field, and the text manuscript field is the content of the explanation process;

[0055] The voice broadcast unit is used to voice broadcast the text of the text field in the auxiliary explanation content.

[0056] In a third aspect, an electronic device is provided, comprising: a memory and a processor;

[0057] The memory is used to store programs;

[0058] The processor is used to execute the program to implement the various steps of the test question auxiliary learning and explanation method described in any one of the first aspects of the present application.

[0059] In a fourth aspect, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the test question auxiliary learning and explanation method described in any one of the aforementioned first aspects of the present application are implemented.

[0060] In a fifth aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the various steps of the test question auxiliary learning and explanation method described in any one of the aforementioned first aspects of the present application.

[0061] By means of the above technical scheme, the present application first obtains the auxiliary learning background information required for the auxiliary learning explanation of the target test questions to be tutored, such as the reference answers of the target test questions, the subject knowledge required for the explanation, the explanation step planning, etc. Under the guidance of the auxiliary learning background information, the ability of the auxiliary learning large model can be called, and the auxiliary learning explanation content of the target test questions can be generated in the form of semantic unit blocks in combination with the auxiliary learning background information and the field meaning of the configured semantic unit block. Among them, the semantic unit block is the smallest semantic unit that independently carries and conveys a complete semantic fragment. The semantic unit block contains a text manuscript field, and the text manuscript field is the content to be broadcast during the explanation process. For the text manuscript in the auxiliary learning explanation content generated by the large model, it is output in the form of voice broadcast, simulating the oral content of the teacher during class, leading students to answer the test questions step by step, and improving the tutoring effect on students. This application does not simply output the reference answers to the test questions, but calls the large model capability to generate auxiliary learning explanation content on the basis of obtaining the background information required for auxiliary learning, which can better guide students to understand the test questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0063] Figure 1 A schematic diagram of an implementation system architecture of the test question assistance explanation method provided in an embodiment of the present application;

[0064] Figure 2 A flowchart of a test question assistance explanation method provided in an embodiment of the present application;

[0065] Figure 3 A schematic diagram of another method for assisting students in learning and explaining test questions provided in an embodiment of the present application;

[0066] Figure 4 A schematic diagram of the structure of a test question assistance and explanation device provided in an embodiment of the present application;

[0067] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0069] This application provides a method for assisting students in learning and explaining test questions, which can be applied to test questions in various subjects. For example, subjects such as Chinese, mathematics, and history. Figure 1 The system architecture shown in FIG. 1 may include a terminal 100 and a server 200. The server 200 may include one or more servers ( Figure 1 A server is included as an example for explanation).

[0070] The terminal 100 or the server 200 can be used alone to execute the test question auxiliary learning explanation method provided in the embodiment of the present application. In addition, the terminal 100 and the server 200 can also be used in collaboration to execute the test question auxiliary learning explanation method provided in the embodiment of the present application.

[0071] Next describe Figure 1 The product form of the mid-terminal 100;

[0072] The terminal 100 in the embodiment of the present application can be a mobile phone, a tablet computer, a learning machine, a teaching screen, a wearable device, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc., and the embodiment of the present application does not impose any restrictions on this.

[0073] The present application embodiment provides a method for assisting learning and explaining test questions, and takes the method as an example of applying the method to a computer device, which can be specifically Figure 1 The terminal 100 or the system consisting of the terminal 100 and the server 200. Figure 2 The test question supplementary explanation method specifically includes the following steps:

[0074] Step S100, obtaining the supplementary learning background information required for supplementary learning explanation of the target test questions.

[0075] The target test questions are the test questions to be supplemented. Typical application scenarios include but are not limited to the following scenarios: when a user practices test questions and finds that he / she does not know how to answer a test question during or after the practice, he / she can record the target test questions in different forms such as text or taking photos, upload them to the test question supplementary learning explanation system of this application, and start supplementary learning for the target test questions.

[0076] The supplementary learning background information corresponding to the target test questions is the background information required for reference in the supplementary learning explanation process.

[0077] In a possible example, the supplementary learning background information may include one or more of the following combinations:

[0078] Reference answers to target test questions, subject knowledge required for explanation, and explanation step planning.

[0079] When the supplementary learning background information includes reference answers, it can help improve the correctness of the answers to the supplementary learning explanations generated subsequently.

[0080] In addition, considering that the auxiliary learning explanation process may also need to use some subject knowledge information, such as professional knowledge of the Chinese language subject, including but not limited to language knowledge (pinyin, characters, vocabulary, punctuation, grammar, etc.), literary common sense knowledge (writers' works, ancient poems, literary and cultural common sense, etc.), reading comprehension knowledge (rhetoric, structure, techniques, emotions, etc.), writing knowledge (writing skills, requirements for practical writing, etc.), etc. When the auxiliary learning background information includes the subject knowledge required for the explanation, a more accurate context is provided for the auxiliary learning model. Through the in-context learning ability of the auxiliary learning model, a more professional explanation is generated, the risk of knowledge errors is reduced, and the probability of "hallucination" phenomenon is reduced. The auxiliary learning model is enabled to have the professional knowledge background for explaining subject test questions.

[0081] In a possible implementation, in order to facilitate students to understand the problem-solving ideas step by step during the auxiliary explanation process, the explanation process can also follow the Socratic guidance form to explain in steps. Therefore, the auxiliary learning background information can include explanation step planning information, that is, to preliminarily split the test question explanation process, making the explanation process easier to understand.

[0082] Step S110, calling the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block to generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0083] The semantic unit block at least includes a text script field, and the text script field is the content of the explanation process.

[0084] A semantic unit block is the smallest semantic unit that independently carries and conveys a complete semantic segment. The explanation content of a semantic unit block can constitute the basic particle of semantic segmentation. A semantic unit block may include more than one field, and different fields have different definitions, so that one or more auxiliary learning contents of different modalities can be generated in the form of a semantic unit block. In this embodiment, the semantic unit block includes at least a text manuscript field (defined as cont for example), which is the content broadcast during the explanation process. Taking Chinese as an example, for the content read word by word during the explanation process, it can generally be set that the text manuscript cannot contain characters that cannot be read directly.

[0085] The auxiliary learning big model is a big model used to generate auxiliary learning explanation content. It can be a general big model or a subject big model trained with subject knowledge. Alternatively, it is also possible to pre-build training data according to the task mode of the auxiliary learning big model, train the big model base, train the big model's ability to generate auxiliary learning explanation content according to the specified template, and obtain the trained big model as the auxiliary learning big model.

[0086] In this step, the auxiliary learning model can generate auxiliary learning explanation content for the target test question in the form of semantic unit blocks under the guidance of auxiliary learning background information and in combination with the field definition of the configured semantic unit blocks, that is, the auxiliary learning explanation content is composed of several semantic unit blocks. The text manuscripts of each semantic unit block can be completely connected.

[0087] Step S120: voice broadcast the text in the text field in the supplementary explanation content.

[0088] Specifically, the text transcripts of each semantic unit block in the generated auxiliary teaching content can be output in the form of voice broadcast, simulating the oral content of a real teacher during class, and leading students to answer questions step by step.

[0089] The test question auxiliary learning explanation method provided in the embodiment of the present application first obtains the auxiliary learning background information required for the auxiliary learning explanation of the target test question to be tutored, such as the reference answer of the target test question, the subject knowledge required for the explanation, the explanation step planning, etc. Under the guidance of the auxiliary learning background information, the ability of the auxiliary learning big model can be called, and the auxiliary learning explanation content of the target test question can be generated in the form of semantic unit blocks in combination with the auxiliary learning background information and the field meaning of the configured semantic unit block. Among them, the semantic unit block is the smallest semantic unit that independently carries and conveys a complete semantic fragment. The semantic unit block contains a text manuscript field, and the text manuscript field is the content to be broadcast during the explanation process. For the text manuscript in the auxiliary learning explanation content generated by the big model, it is output in the form of voice broadcast, simulating the oral content of the teacher during class, leading students to answer the test questions step by step, and improving the tutoring effect on students. This application is not a simple output of the reference answer to the test question, but on the basis of obtaining the background information required for auxiliary learning, calling the big model ability to generate auxiliary learning explanation content, which can better guide students to understand the test questions.

[0090] As mentioned above, the supplementary learning background information may include one or more of the reference answers to the target test questions, the subject knowledge required for the explanation, and the explanation step planning information.

[0091] Next, the process of obtaining each type of supplementary learning background information is introduced respectively.

[0092] 1. The process of obtaining reference answers to target test questions

[0093] In a possible implementation, it is possible to query whether the answer information of the target test question exists in the configured test question library. If so, the reference answer corresponding to the target test question in the test question library is directly obtained.

[0094] In another possible implementation, a reference answer to the target question may be generated. For example, when the answer information of the target question does not exist in the configured question database, the reference answer to the target question may be obtained by actively generating the answer.

[0095] This embodiment provides a method for generating reference answers to target test questions, and can call a large model to instruct the large model to generate reference answers to target test questions. With the help of the natural language understanding and generation capabilities of the large model, reference answers to target test questions can be generated.

[0096] The big model can be a general big model or a subject big model (i.e., a big model trained with subject knowledge). In addition, the answer data can be used to train the big model base to obtain a trained big model, thereby improving the ability of the big model to generate reference answers for target test questions. The answer data includes question data marked with correct answers.

[0097] This embodiment provides another method for generating a reference answer to a target test question, specifically:

[0098] First, similar question data that meets the similarity requirements with the target test question is retrieved.

[0099] Further, the big model is called to instruct the big model to refer to similar question data and generate reference answers to the target test questions.

[0100] Among them, similar question data refers to question data with a high degree of similarity to the target question, which may include question stems, answers, analytical information, etc. By retrieving similar question data, it can be provided to the big model to assist the big model in generating more accurate reference answers to the target question.

[0101] The process of retrieving similar question data that meets the similarity requirement with the target question can be performed in a variety of data sources, such as on the Internet, in a pre-established question bank, etc. In this embodiment, the retrieval process in the question bank is taken as an example for description.

[0102] During retrieval, either a sparse vector recall strategy or a dense vector recall strategy or a combination of the two can be used.

[0103] Among them, sparse vectors refer to those constructed using methods such as Bag of Words and TF-IDF (Term Frequency - Inverse Document Frequency). Taking TF-IDF as an example, this method calculates a weight for each word in a document, reflecting the importance of the word in the document and its rarity in the entire corpus. Based on the dimension of the question bank, each question is regarded as a document and its TF-IDF vector is calculated.

[0104] Dense vectors are generated using pre-trained language models (such as BERT). These models can capture the semantic information of the text and map the test text into a low-dimensional vector space. For example, the pre-trained BERT model is used to encode the test questions to obtain a dense vector representation of each test question.

[0105] For the target test question, the sparse vector model and the dense vector model can be used to generate the vector (sparse vector and / or dense vector) of this test question. Similarly, the vector (sparse vector and / or dense vector) is calculated for each test question in the pre-configured test question bank. By calculating the similarity (such as cosine similarity) between the vector of the target test question and the vector of each test question in the test question bank, and sorting them from high to low, the top several test question data are selected, and similar question data is recalled from the test question bank.

[0106] Further optionally, after recalling similar question data, a process of re-ranking the similar question data can be further added, and the recalled similar question data can be precisely ranked through the re-ranking model, so as to determine several (for example, 1 to 2) question data that are most similar to the target question as the final result.

[0107] Specifically, in the above steps, the vector recall method can roughly find test questions with similar content, but if you want to improve the matching effect of similar questions in a specific way, you can use the re-ranking model to further screen the rough sorting results. For example, you can extract test question features from dimensions such as text features (such as vocabulary overlap, semantic similarity), test question quality features (such as the difficulty coefficient of the test questions, error rate, completeness of the test question answers, etc.), and test point features (such as the knowledge points involved, the importance of the test points, the examination method, the question type, the question category, etc.), and then use statistical methods, machine learning algorithms or deep learning models to learn the comprehensive score ranking of the rough sorting results, and select several similar question results with the highest scores to return. In the re-ranking process, the re-ranking model will comprehensively consider various features, not only focusing on the similarity of the text, but also combining information such as test question quality and test points, so as to obtain more accurate and valuable similar question data.

[0108] Since the size of the question bank is necessarily exhaustible, its timeliness and accuracy cannot be fully guaranteed. Therefore, this application combines the generation capabilities of a large model, and the large model generates reference answers to the target questions by referring to similar question data, further enhancing the accuracy of answer acquisition.

[0109] In this embodiment, a small amount of carefully processed test questions can be used to train the large automatic answering model first, so as to stimulate the large model's ability to use the underlying massive knowledge to answer subject questions. Then, similar question data containing question stems, answers, and analytical information can be used as reference information for the large model input, which can further enhance the large model's answering effect. The solution provided in this embodiment that combines similar question data retrieval and large model answer generation can combine the advantages of both, improve the accuracy of the reference answers to the generated target test questions, and lay a more reliable premise for the subsequent auxiliary learning process.

[0110] Abstract representation, definition The reference answer to the target test question is obtained by splicing the automatic answer prompt words , similar question data , Target Questions , using the automatic answering model Generate reference answers:

[0111] A a uto-ans = LLM a uto-ans ( [p auto -ans , Q similar , Q ]) 。

[0112] 2. The acquisition process of the subject knowledge required for the explanation of target test questions

[0113] In general, during pre-training and subsequent training, the tutoring large model can learn relevant knowledge content from a vast amount of text. However, due to the mixture of correct and incorrect knowledge, and the uneven distribution of subject knowledge in the text, the degree to which the model can learn cannot be guaranteed, which may lead to the model not always following correct subject knowledge during the generation process.

[0114] Therefore, this application can obtain the subject knowledge information required for the explanation of target test questions and add it to the tutoring background information, thereby guiding the tutoring large model to generate more accurate tutoring explanation content.

[0115] This application can pre-construct a high-quality subject knowledge base by professional teaching and research teachers and process and store it in text form. At the same time, a retrieval condition generation model is trained using sample test questions labeled with subject knowledge retrieval condition tags. When in use, the target test question is input into the retrieval condition generation model, and the subject knowledge retrieval condition corresponding to the target test question generated by the model can be obtained. Further, the generated subject knowledge retrieval condition can be used to retrieve the target subject knowledge in the subject knowledge base as the subject knowledge required during the explanation process of the target test question.

[0116] Taking the target test questions in the Chinese subject as an example for illustration:

[0117] Target test question 1 is: When using the radical search method for the character '尽', it should be searched under the __ radical and then search for __ more strokes.

[0118] Target test question 2 is: Please judge whether the following statement is correct: When a cricket builds a house, it first uses an existing cave and then excavates and transforms it. (True or False).

[0119] For target test question 1, through the retrieval condition generation model, the subject knowledge retrieval conditions can be obtained: ① Query purpose: Radical; Query input: 尽. ② Query purpose: Strokes; Query input: 尽.

[0120] For target test question 2, through the retrieval condition generation model, the subject knowledge retrieval condition can be obtained: ③ Query purpose: Text, Query input: When a cricket builds a house, it first uses an existing cave and then excavates and transforms it.

[0121] For the process of retrieving the target subject knowledge in the subject knowledge base using the subject knowledge retrieval condition, this embodiment provides two optional retrieval strategies:

[0122] First, if the subject knowledge retrieval condition conforms to the template rule, a heuristic rule retrieval strategy can be adopted to retrieve the target subject knowledge in the subject knowledge base.

[0123] Taking the subject knowledge retrieval condition ① in the above example as an example, since it conforms to the template rule, according to the heuristic rule detection strategy, the radical of the character "尽" can be searched in the configured "radical" knowledge base, and the query result is that the radical of the character "尽" is "尸".

[0124] Second, if the subject knowledge retrieval condition does not conform to the template rule, a vector retrieval strategy can be adopted to retrieve the target subject knowledge in the subject knowledge base.

[0125] Specifically, the query vector representation corresponding to the subject knowledge retrieval condition can be determined, and based on the vector similarity, the target subject knowledge that meets the similarity requirement with the query vector representation can be determined in the subject knowledge base.

[0126] Taking the subject knowledge retrieval condition ③ in the above example as an example, since it does not conform to the template rule, the vector retrieval strategy is used. First, the query input in the subject knowledge retrieval condition is vectorized, and then in the "text" knowledge base, similar texts are searched through vector similarity calculation as the target subject knowledge.

[0127] For the above retrieval condition generation model, it can be various types of generative models. In a possible example, the retrieval condition generation model can be a retrieval condition generation large model obtained by training an initial large model with sample test questions labeled with subject knowledge retrieval condition tags. The ability of the large model can improve the correctness of the generated subject knowledge retrieval conditions.

[0128] In a possible implementation, the process of introducing the call to the retrieval condition generation large model to generate the subject knowledge retrieval condition corresponding to the target test question is specifically as follows:

[0129] Call the retrieval condition generation large model to instruct the large model to select the target intention to which the target test question belongs and generate the parameter information of the target test question under the target intention in combination with the configured intention list and parameter definition, where the intention list contains the intention of each knowledge point inspection under the subject to which the target test question belongs.

[0130] The subject knowledge retrieval condition corresponding to the target test question is composed of the target intention and the corresponding parameter information.

[0131] Specifically, the present application can pre-organize the intention list and the parameter definition corresponding to each intention according to the intention of each knowledge point inspection under the subject to which the target test question belongs. The parameter corresponding to the intention represents the parameter that needs to be filled in for the current intention.

[0132] Take an intent under the Chinese language subject as an example: "Query purpose: [], query input: []". The [] contain the specific parameter values ​​that need to be filled in the intent. For the target test question 1 in the above example, the query purpose parameter is [radical], and the query input parameter is [end].

[0133] In this embodiment, by defining the intent list and parameter definition, the retrieval condition generation model can be guided to select the target intent to which the target test question belongs from the intent list, and generate parameter information of the target test question under the target intent. Finally, the subject knowledge retrieval condition corresponding to the target test question is composed of the target intent and the corresponding parameter information.

[0134] It can be understood that the number of target intentions to which the target test question belongs can be one or more, that is, the subject knowledge retrieval conditions corresponding to the target test question finally generated can be one or more.

[0135] By pre-defining the intent list, the retrieval condition generation model can focus on the target intent of selecting the target test question in the intent list, avoiding the generation of erroneous retrieval conditions due to the large model hallucination problem, and improving the accuracy of the subject knowledge retrieval conditions generated by the retrieval condition generation model.

[0136] Abstract representation, definition In order to explain the subject knowledge required for the target test questions, prompt words are generated by splicing knowledge search conditions , use the search condition generation model to generate subject knowledge search conditions . Then use the knowledge retrieval system , retrieve the required specific subject knowledge:

[0137] Query know ledge = LLM knowledge ( [p knowledge , Q ])

[0138] .

[0139] 3. The process of obtaining information on the steps and rules of explaining the target test questions

[0140] In teaching scenarios, questions facing different test points and examination methods generally have different logic of explanation steps. Teachers can precipitate explanation methods that are easy for students to accept in long-term teaching activities, but generally large language models do not have teaching experience, and the explanation process may appear stiff and unclear. For this reason, this application can add the explanation step planning information of the target test questions to the auxiliary learning background information, provide guidance on the explanation steps for the auxiliary learning large model to generate auxiliary learning explanation content, improve the quality of the generated auxiliary learning explanation content, and make it easier for students to accept.

[0141] In this embodiment, the big model capability can be used to generate explanation step planning information of the target test questions using the big model through auxiliary learning steps.

[0142] The large model can be trained in advance using question data labeled with explanation step planning information. By using a small amount of high-quality step planning data to train the large model, the large model can plan reasonable explanation steps for the test questions like a professional teacher.

[0143] The input of the auxiliary learning step generation model is the target test question, and the expected output is the explanation steps that must be relied on in the process of explaining this question. The large model mainly breaks down the test question-solving process in detail, learns from the training data the key points that real teachers often emphasize when explaining, and provides methodological guidance for the overall explanation.

[0144] Typical examples are as follows:

[0145] "Subject:

[0146] Write sentences as required.

[0147] Example: Do we need to treat the bugs on the leaves? There is no need to treat the bugs on the leaves.

[0148] You are so old now, do you still need your mother to take you to school?

[0149] _________________________".

[0150] The explanation step planning information of the above test questions generated by the large model generated through the auxiliary learning steps is as follows:

[0151] This question tests how to convert a rhetorical question into a declarative sentence.

[0152] Step 1: Analyze the requirements of the question and introduce the solution method.

[0153] Step 2: Add or delete words that express interrogative tone.

[0154] Step 3: Compare the meaning of the original sentence and add or delete negative words.

[0155] Step 4: Modify punctuation.

[0156] Step 5: Answer the questions as required.

[0157] Finally, the answers to the test questions are displayed, and the answer ideas are summarized and reviewed.

[0158] The above process of generating the explanation step planning information of the target test questions through the auxiliary learning steps is abstractly represented as follows:

[0159] definition To explain the steps to be followed in the target test questions, we need to plan information and generate prompt words by splicing the auxiliary learning steps. and the target test content Q, using the auxiliary learning steps to generate a large model Generate explanation step planning information:

[0160] BI Steps = LLM s teps ( [p Steps , Q ]) .

[0161] In some embodiments of the present application, step S110 is further described to call the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block to generate the auxiliary learning explanation content of the target test question in the form of semantic unit blocks.

[0162] The above-mentioned embodiment introduces that the semantic unit block includes at least a text script field, which is the content that needs to be broadcasted during the auxiliary teaching process. After obtaining the auxiliary teaching content, the text script content can be broadcasted to achieve the output of the auxiliary teaching content in audio mode, simulating the oral content of the teacher's real class explanation of the test questions.

[0163] On this basis, this embodiment further provides a multi-modal auxiliary learning scenario. That is, in addition to the text manuscript field, the semantic unit block can also include other fields, so as to facilitate the subsequent output of corresponding auxiliary learning explanation content according to other modes.

[0164] Refer to Table 1 below, which illustrates an optional field definition of a semantic unit block:

[0165] Table 1

[0166]

[0167] It should be noted that the above Table 1 only illustrates an optional field definition method of the semantic unit block. In actual application, a combination of one or more fields can be selected from the above fields, or other fields can be added.

[0168] In a possible implementation, the semantic unit block may include the display content field display_cont.

[0169] The auxiliary learning explanation content of the target test question generated by the auxiliary learning model is called, and the blackboard content corresponding to the display content field is also included. On this basis, the test question auxiliary learning explanation method of this application can also include the following steps:

[0170] Based on the voice broadcast transcript, the blackboard content corresponding to the display content field will be displayed on the screen.

[0171] Specifically, the auxiliary learning explanation content generated by the auxiliary learning model includes semantic unit blocks that are output one by one. The semantic unit block can include both the text manuscript field and the display content field. The value of the display content field is the blackboard content that is displayed synchronously with the text manuscript content of the text manuscript field. It is bound one by one with the text manuscript content. The blackboard content can be displayed on the screen while the text manuscript content is broadcast. It is concise but more written, simulating the blackboard writing of a real teacher during class, helping students to better understand and remember.

[0172] Optionally, the blackboard content of the content field displayed in this embodiment can support the display of key fragments in the form of rich text.

[0173] The test question auxiliary learning and explanation method provided in this embodiment can realize the audio playback of the text manuscript and the image display of the blackboard content, that is, it can realize the output of auxiliary learning and explanation content in multimodal form, which is easier to help students understand.

[0174] In another possible implementation, the semantic unit block may include the above-mentioned test question marking content field mark_cont.

[0175] The auxiliary learning explanation content of the target test question generated by the auxiliary learning model is called, and the test question fragment corresponding to the test question marking content field is also included. On this basis, the test question auxiliary learning explanation method of the present application can also include the following steps:

[0176] Based on the voice broadcast transcript, the test question segment corresponding to the test question marking content field in the target test question is marked and displayed.

[0177] Specifically, the auxiliary learning explanation content generated by the auxiliary learning model includes semantic unit blocks that are output one by one. The semantic unit block can include both the text manuscript field and the test question marking content field. The value of the test question marking content field is the test question segment that needs to be marked in synchronization with the text manuscript content of the text manuscript field. It is bound one by one with the text manuscript content, and the test question segment corresponding to the test question marking content field in the target test question can be marked and displayed on the screen while the text manuscript content is broadcast. This simulates the process of a real teacher explaining in class, and it is often necessary to mark out key information such as the key words in the test question stem and the key sentences in the original question materials to enhance students' attention.

[0178] In a possible usage scenario, students can start the test question auxiliary learning and explanation process by uploading the target test question image. In this embodiment, the test question segment in the test question marking content field can be marked in the image uploaded by the student.

[0179] When marking the test question fragments, the default marking form can be used, such as underlining, highlighting, etc. In addition, the test question marking content field in the semantic unit block can also include the marking form, that is, when the auxiliary learning model generates the test question fragments that need to be marked, it can further determine the marking form of the test question fragments. On this basis, the test question fragments corresponding to the test question marking content field in the target test question can be marked and displayed according to the corresponding marking form.

[0180] As shown in Table 1 above, the semantic unit block may also include the current explanation phase field step. That is, the application may predefine the test question explanation phase, for example, including four phases: "reading phase", "analysis phase", "answer phase", and "summary and expansion phase".

[0181] The semantic unit block may also include the current semantic unit block type field type. The value may include "lecture", "choice question", "heuristic question", "knowledge transfer", "discussion dialogue", etc., which may be set according to the business scenario needs.

[0182] The semantic unit block may also include a role field, and the role value is "teacher".

[0183] In a possible implementation, when the value of the semantic unit block type field is "discuss dialogue", the value of the display content field display_cont may be set to be empty.

[0184] In a possible implementation, it can be set that only when the value of the current explanation link field is "analysis link", the value of the test question mark content field mark_cont is not empty, and when the value of the current explanation link field is not "analysis link", the value of the test question mark content field mark_cont is empty.

[0185] It should be noted that the semantic unit block may include a combination of one or more fields in Table 1 above. In one possible implementation, the semantic unit block may simultaneously include: a text field, a display content field, and a test question marking content field, thereby enabling the output of auxiliary learning explanation content in a multimodal form. By conducting the auxiliary learning explanation process in a multimodal form, it is possible to simulate the real teacher's teaching process and improve the quality of auxiliary learning explanations.

[0186] In some possible implementations, the test question supplementary explanation method of the present application can support the interaction between the machine and the user. That is, the user can provide feedback on the supplementary explanation content generated by the supplementary learning model each time, and the supplementary learning model will continue to generate subsequent supplementary explanation content according to the user feedback content until the explanation is completed.

[0187] Therefore, the test question auxiliary learning explanation method of the present application may also include the following steps:

[0188] Obtain user feedback on the output supplementary teaching content.

[0189] The auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field meaning of the configured semantic unit block and the feedback content, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0190] Refer to Table 2 below, which illustrates the possible interactions between different roles during the test question explanation process.

[0191] Table 2

[0192]

[0193] Abstract representation, definition To explain the content of the auxiliary learning, integrate various types of auxiliary learning background information ( , , ), use learning aids Trigger auxiliary learning model The auxiliary learning ability generates semantic unit blocks. During the explanation process, there is interaction with the user, so the model input may contain user reply content. , the model will respond accordingly based on the user's reply:

[0194] R tutor = LLM tutor ( [p tutor ,[ A a uto-ans , BI knowledge , BI Steps , C U , Q ]) .

[0195] Reference Figure 3 , which illustrates a flowchart of another method of auxiliary explanation of test questions.

[0196] The entire test question auxiliary learning explanation framework can include: an agent scheduling layer, and several auxiliary learning background information acquisition modules, such as: a reference answer agent module, a subject knowledge agent module, a step planning agent module, a response generation module, and a self-reflection module.

[0197] For the target test questions uploaded by users to be assisted in learning, each auxiliary learning background information acquisition module can be accessed through the agent scheduling layer, such as obtaining the reference answers of the target test questions through the reference answer agent module, obtaining the subject knowledge required for explanation through the subject knowledge agent module, and obtaining the explanation step planning information through the step planning agent module.

[0198] Among them, the reference answer proxy module can recall similar question data from the question bank through sparse plus dense vector coarse recall, combined with the reordering and fine sorting strategy, and then call the large model to generate reference answers for the target questions based on similar question data. The detailed process can be referred to the previous introduction.

[0199] The subject knowledge agent module can use the large model to generate the subject knowledge retrieval conditions of the target test questions, and then adopt the heuristic rule retrieval strategy or the vector retrieval strategy to retrieve the subject knowledge from the subject knowledge base. The detailed process can be referred to the previous introduction.

[0200] The step planning agent module can use the large model to generate the explanation step planning information of the target test questions. The detailed process can be referred to the previous introduction.

[0201] The proxy scheduling layer sends the acquired supplementary learning background information to the reply generation module. In addition, this application supports human-computer interaction in the supplementary learning process, so the user's multimodal input information can be simultaneously sent to the reply generation module, and the reply generation module calls the supplementary learning model to generate supplementary learning explanation content, which can be specifically output in the form of streaming semantic unit blocks. According to the content of different fields in the semantic unit block, multimodal output is adopted, and please refer to the previous introduction for details.

[0202] Through the reply generation module, the system has been able to initially output relatively reasonable and understandable supplementary explanation content. However, there may still be explanation problems such as model knowledge errors and hallucinations.

[0203] In a possible implementation, the present embodiment can further introduce a self-reflection module to check the supplementary learning explanation content generated by the supplementary learning model, propose modification suggestions, and further improve the supplementary learning effect. Finally, the reply generation module is used to generate the reply, and it is rendered and presented in a multimodal form.

[0204] The test question supplementary explanation method of this embodiment may further include:

[0205] Evaluate the auxiliary learning explanation content generated by the auxiliary learning model and generate modification suggestions.

[0206] When the evaluation result indicates that the supplementary explanation content generated this time meets the requirements, the supplementary explanation content is output. The process of outputting the supplementary explanation content may include output of one or more modes, which specifically needs to be based on the field information contained in the semantic unit block in the supplementary explanation content. For example, if the semantic unit block includes a text manuscript field, the process of outputting the supplementary explanation content may be to execute the aforementioned step S120 and voice broadcast the text manuscript of the text manuscript field in the supplementary explanation content.

[0207] The evaluation result of the supplementary teaching content can indicate whether the quality of the supplementary teaching content meets the requirements. For example, the evaluation result can be: qualified / unqualified, or: excellent / good / poor, or it can be a quality score value.

[0208] Further optionally, when the evaluation results indicate that the supplementary learning explanation content generated this time does not meet the requirements, the supplementary learning big model can be called to instruct the supplementary learning big model to combine the supplementary learning background information, the field definition of the configured semantic unit block, the supplementary learning explanation content generated this time, the evaluation results and the modification suggestions, and regenerate the supplementary learning explanation content of the target test question in the form of semantic unit blocks.

[0209] In this embodiment, by evaluating the results generated by the auxiliary learning big model and generating modification suggestions, when the evaluation results indicate that the quality does not meet the requirements, the auxiliary learning big model can be guided to regenerate the auxiliary learning explanation content, and the auxiliary learning big model can be guided to generate more accurate and useful explanation content, thereby further improving the reliability of the generation results of the auxiliary learning big model.

[0210] In a possible implementation, the process of evaluating the auxiliary learning explanation content generated by the auxiliary learning big model and generating modification suggestions can be implemented by the big model, which is defined as a self-reflection big model. The self-reflection big model can be called to instruct the big model to evaluate the auxiliary learning explanation content generated this time and generate modification suggestions.

[0211] By utilizing the self-checking capability of the self-reflective big model, the reliability of the results generated by the auxiliary learning big model can be further improved.

[0212] The abstract representation is defined as To evaluate the results of the supplementary teaching content generated this time, For revision suggestions. Self-reflection model Use self-reflection prompts , various types of supplementary learning background information, supplementary learning explanation content generated this time and user response content Make joint decisions and give evaluation results and modification suggestions:

[0213] E tutor , S tutor = LLM self -critic ( [p self-critic ,[ A a uto-ans , BI knowledge , BI Steps , C U , Q , R tutor ])

[0214] Setting the regeneration condition threshold If the evaluation result is higher than the threshold value T, it means that the quality of the auxiliary learning explanation content generated this time is qualified and can be directly output; if the evaluation result is not higher than the threshold value T, it means that the quality of the auxiliary learning explanation content generated this time is unqualified and needs to be comprehensively modified. The auxiliary learning model regenerates the final auxiliary learning explanation content. , the formula is as follows:

[0215]

[0216] .

[0217] In a possible implementation, the self-reflection module can evaluate the generated supplementary teaching content according to the configured evaluation dimensions. The configured evaluation dimensions include but are not limited to:

[0218] Whether the evaluation of students' answers is accurate, whether there are knowledge errors in the explanation, whether there are logical errors in the explanation, and whether the explanation is useful (i.e., whether the explanation can promote students' understanding of the solution ideas of the target test questions).

[0219] Through the evaluation dimensions of the above examples, the self-reflection module can evaluate the auxiliary teaching content generated this time and put forward specific modification suggestions. For example:

[0220] "①Accurate evaluation; ②There are knowledge errors: After removing the radical of the character '尽', the remaining 4 strokes are incorrectly described. It should be 'a right-falling stroke, a dot, a dot', a total of 3 strokes; ③There are no logical errors; ④The explanation is highly useful."

[0221] The self-reflection module plays the role of a professional reviewer. The short but clear evaluation and modification opinions proposed are easier to generate, but can play a clear guiding role in the output of the response generation module. The evaluation results and modification opinions are given to the response generation module again. Through this self-reflection process, it guides the generation of more accurate and useful supplementary learning explanation content.

[0222] The following describes the test question supplementary learning explanation device provided by the embodiments of the present application. The test question supplementary learning explanation device described below can be correspondingly referred to the test question supplementary learning explanation method described above.

[0223] See Figure 4 , Figure 4 which is a schematic structural diagram of a test question supplementary learning explanation device disclosed in the embodiments of the present application.

[0224] As Figure 4 shown, the device may include:

[0225] The supplementary learning background information acquisition unit 11 is used to acquire the supplementary learning background information required for supplementary learning explanation of the target test question. The supplementary learning background information includes: the reference answer of the target test question, the subject knowledge required for the explanation, and the explanation step plan;

[0226] The supplementary learning explanation content generation unit 12 is used to call the supplementary learning large model to instruct the supplementary learning large model to generate the supplementary learning explanation content of the target test question in the form of the semantic unit block in combination with the supplementary learning background information and the field definition of the configured semantic unit block. Among them, the semantic unit block at least includes a transcript field, and the transcript field is the content broadcast during the explanation process;

[0227] The voice broadcast unit 13 is used to perform voice broadcast on the transcript of the transcript field in the supplementary learning explanation content.

[0228] In a possible implementation, the supplementary learning background information includes one or more of the following combinations:

[0229] The reference answer of the target test question, the subject knowledge required for the explanation, the explanation step plan information.

[0230] In a possible implementation, the process of the supplementary learning background information acquisition unit acquiring the reference answer of the target test question includes:

[0231] Calling the large model to instruct the large model to generate the reference answer of the target test question.

[0232] In a possible implementation, the process of the auxiliary learning background information acquisition unit acquiring the reference answer to the target test question includes:

[0233] Retrieving similar question data that meets similarity requirements with the target test question;

[0234] The big model is called to instruct the big model to refer to the similar question data and generate a reference answer to the target test question.

[0235] In a possible implementation, the process of the auxiliary learning background information acquisition unit acquiring the subject knowledge required for explaining the target test question includes:

[0236] Calling the configured retrieval condition generation model to generate the subject knowledge retrieval condition corresponding to the target test question, wherein the retrieval condition generation model is trained using sample test questions annotated with subject knowledge retrieval condition labels;

[0237] The subject knowledge search conditions are used to search for target subject knowledge in a subject knowledge base as the subject knowledge required for the target test question explanation process.

[0238] In a possible implementation, the retrieval condition generation model is a retrieval condition generation model obtained by training the initial large model with sample test questions labeled with subject knowledge retrieval condition labels. On this basis, the auxiliary learning background information acquisition unit calls the retrieval condition generation model to generate the subject knowledge retrieval condition corresponding to the target test question, including:

[0239] Calling the retrieval condition to generate a large model to instruct the large model to combine the configured intent list and parameter definition, select the target intent to which the target test question belongs and generate parameter information of the target test question under the target intent, wherein the intent list includes the examination intent of each knowledge point under the subject to which the target test question belongs;

[0240] The target intention and the corresponding parameter information constitute the subject knowledge retrieval condition corresponding to the target test question.

[0241] In a possible implementation, the process of the auxiliary learning background information acquisition unit using the subject knowledge search condition to search for target subject knowledge in the subject knowledge base includes:

[0242] If the subject knowledge search condition meets the template rule, the heuristic rule search strategy is used to search for the target subject knowledge in the subject knowledge base;

[0243] If the subject knowledge retrieval condition does not meet the template rule, the query vector representation corresponding to the subject knowledge retrieval condition is determined, and based on the vector similarity, the target subject knowledge that meets the similarity requirement with the query vector representation is determined in the subject knowledge base.

[0244] In a possible implementation, the process of the auxiliary learning background information acquisition unit acquiring the explanation step planning information of the target test question includes:

[0245] The big model is called to instruct the big model to generate explanation step planning information for the target test question.

[0246] In a possible implementation, the semantic unit block further includes: a display content field, and the display content field is the blackboard content on the upper screen. Then the auxiliary learning explanation content generation unit calls the auxiliary learning big model to generate the auxiliary learning explanation content of the target test question, and also includes the blackboard content corresponding to the display content field. The device of the present application may also include:

[0247] The blackboard content display unit is used to display the blackboard content corresponding to the display content field on the screen based on the voice broadcast of the text manuscript by the voice broadcast unit.

[0248] In a possible implementation, the semantic unit block further includes: a test question marking content field, wherein the test question marking content field is a test question segment that needs to be marked when explaining the current content. Then the auxiliary learning explanation content generation unit calls the auxiliary learning big model to generate the target test question, and also includes the test question segment corresponding to the test question marking content field. The device of the present application may also include:

[0249] The test question marking unit is used to mark and display the test question segment corresponding to the test question marking content field in the target test question based on the voice broadcast of the text script by the voice broadcast unit.

[0250] In a possible implementation, the question marking content field also includes a marking form, and the question marking unit displays the question segment corresponding to the question marking content field in the target question by marking the question, including:

[0251] The test question segment corresponding to the test question marking content field in the target test question is marked and displayed in a corresponding marking format.

[0252] In a possible implementation, the device of the present application may further include:

[0253] A human-computer interaction unit, used to obtain user feedback on the output auxiliary teaching content;

[0254] The auxiliary learning explanation content generation unit is also used to: call the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field meaning of the configured semantic unit block and the feedback content, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0255] In a possible implementation, the device of the present application may further include:

[0256] The self-reflection unit is used to evaluate the supplementary learning explanation content generated by the supplementary learning explanation content generation unit and generate modification suggestions;

[0257] When the evaluation result indicates that the supplementary explanation content generated this time meets the requirements, the processing steps of the voice broadcast unit, the blackboard content display unit and / or the test question marking unit are executed.

[0258] In a possible implementation, the auxiliary teaching content generation unit is further used for:

[0259] When the evaluation result generated by the self-reflection unit indicates that the auxiliary learning explanation content generated this time does not meet the requirements, the auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field definition of the configured semantic unit block, the auxiliary learning explanation content generated this time, the evaluation result and the modification suggestions, and regenerate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

[0260] In a possible implementation, the process in which the self-reflection unit evaluates the supplementary teaching content generated this time and generates modification suggestions includes:

[0261] The big model is called to instruct the big model to evaluate the supplementary teaching content generated this time and generate modification suggestions.

[0262] In a possible implementation, the self-reflection unit evaluates the auxiliary teaching content generated this time according to the configured evaluation dimensions, and the evaluation dimensions include at least one of the following:

[0263] Whether the evaluation of students' answers is accurate, whether there are knowledge errors in the explanation, whether there are logical errors in the explanation, and whether the explanation can promote students' understanding of the solution ideas of the target test questions.

[0264] The present application also provides an electronic device in an embodiment. Figure 5 As shown, it shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiment of the present application. The electronic device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, tablet computers, learning machines, teaching large screens, wearable devices, etc. Figure 5 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0265] like Figure 5As shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603, so as to implement the test question auxiliary learning explanation method of the aforementioned embodiment of the present application. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0266] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0267] Also provided in an embodiment of the present application is a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the test question auxiliary learning and explanation methods provided in the embodiments of the present application.

[0268] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any one of the test question auxiliary learning and explanation methods provided in the embodiment of the present application.

[0269] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0270] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0271] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0272] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0273] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

Claims

1. A method for assisting students in learning and explaining test questions, characterized in that: include: Obtain the background information needed to provide supplementary explanations for target test questions; Calling the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block, wherein the semantic unit block at least includes a text script field, and the text script field is the content of the explanation process; The text of the text field in the auxiliary explanation content is voice broadcasted.

2. The method according to claim 1, characterized in that The supplementary learning background information includes one or more of the following combinations: Reference answers to the target test questions, subject knowledge required for explanation, and explanation step planning information.

3. The method according to claim 2, characterized in that The process of obtaining the reference answer of the target test question includes: The big model is called to instruct the big model to generate a reference answer to the target test question.

4. The method according to claim 3, characterized in that Before calling the large model to generate the reference answer of the target test question, it also includes: Retrieving similar question data that meets similarity requirements with the target test question; The process of calling the large model to generate a reference answer to the target test question includes: The big model is called to instruct the big model to refer to the similar question data and generate a reference answer to the target test question.

5. The method according to claim 2, characterized in that: The process of acquiring the subject knowledge required for explaining the target test questions includes: Calling the configured retrieval condition generation model to generate the subject knowledge retrieval condition corresponding to the target test question, wherein the retrieval condition generation model is trained using sample test questions annotated with subject knowledge retrieval condition labels; The subject knowledge search conditions are used to search for target subject knowledge in a subject knowledge base as the subject knowledge required for the target test question explanation process.

6. The method according to claim 5, characterized in that The retrieval condition generation model is a retrieval condition generation model obtained by training the initial large model with sample test questions annotated with subject knowledge retrieval condition labels; The process of calling the retrieval condition generation macro model to generate the subject knowledge retrieval condition corresponding to the target test question includes: Calling the retrieval condition to generate a large model to instruct the large model to combine the configured intent list and parameter definition, select the target intent to which the target test question belongs and generate parameter information of the target test question under the target intent, wherein the intent list includes the examination intent of each knowledge point under the subject to which the target test question belongs; The target intention and the corresponding parameter information constitute the subject knowledge retrieval condition corresponding to the target test question.

7. The method according to claim 5, characterized in that The process of searching the target subject knowledge in the subject knowledge base using the subject knowledge search condition includes: If the subject knowledge search condition meets the template rule, the heuristic rule search strategy is used to search for the target subject knowledge in the subject knowledge base; If the subject knowledge retrieval condition does not meet the template rule, the query vector representation corresponding to the subject knowledge retrieval condition is determined, and based on the vector similarity, the target subject knowledge that meets the similarity requirement with the query vector representation is determined in the subject knowledge base.

8. The method according to claim 2, characterized in that: The process of obtaining the explanation step planning information of the target test question includes: The big model is called to instruct the big model to generate explanation step planning information for the target test question.

9. The method according to claim 1, characterized in that: The semantic unit block also includes: a display content field, the display content field is the blackboard content on the upper screen; Then the auxiliary learning explanation content of the target test question generated by the auxiliary learning big model is called to also include the blackboard content corresponding to the display content field; The method further includes: On the basis of the voice broadcast of the text script, the blackboard content corresponding to the display content field is displayed on the screen.

10. The method according to claim 1, characterized in that The semantic unit block also includes: a test question marking content field, wherein the test question marking content field is a test question segment that needs to be marked when explaining the current content; Then the auxiliary learning explanation content of the target test question generated by the auxiliary learning big model is called to also include the test question fragment corresponding to the test question marking content field; The method further includes: On the basis of the voice broadcast of the text script, the test question segment corresponding to the test question marking content field in the target test question is marked and displayed.

11. The method according to claim 10, characterized in that The test question marking content field also includes a marking form; The process of marking and displaying the test question segment corresponding to the test question marking content field in the target test question includes: The test question segment corresponding to the test question marking content field in the target test question is marked and displayed in a corresponding marking format.

12. The method according to claim 1, characterized in that Also includes: Obtain user feedback on the output supplementary teaching content; The auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field meaning of the configured semantic unit block and the feedback content, and generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

13. The method according to any one of claims 1 to 12, characterized in that: Before the voice broadcast of the text in the text field in the auxiliary teaching content, the method further includes: Evaluate the supplementary teaching content generated this time and generate modification suggestions; When the evaluation result indicates that the supplementary teaching explanation content generated this time meets the requirements, the step of voice broadcasting the text script in the text script field in the supplementary teaching explanation content is executed.

14. The method according to claim 13, characterized in that Also includes: When the evaluation result indicates that the auxiliary learning explanation content generated this time does not meet the requirements, the auxiliary learning big model is called to instruct the auxiliary learning big model to combine the auxiliary learning background information, the field definition of the configured semantic unit block, the auxiliary learning explanation content generated this time, the evaluation result and the modification suggestion, and regenerate the auxiliary learning explanation content of the target test question in the form of the semantic unit block.

15. The method according to claim 13, characterized in that The process of evaluating the generated supplementary teaching content and generating modification suggestions includes: The big model is called to instruct the big model to evaluate the supplementary teaching content generated this time and generate modification suggestions.

16. The method according to claim 13, characterized in that The process of evaluating the supplementary teaching content generated this time is evaluated according to the configured evaluation dimensions, and the evaluation dimensions include at least one of the following: Whether the evaluation of students' answers is accurate, whether there are knowledge errors in the explanation, whether there are logical errors in the explanation, and whether the explanation can promote students' understanding of the solution ideas of the target test questions.

17. A test question auxiliary explanation device, characterized in that: include: A supplementary learning background information acquisition unit is used to acquire supplementary learning background information required for supplementary learning explanation of the target test question, wherein the supplementary learning background information includes: a reference answer to the target test question, subject knowledge required for explanation, and explanation step planning; The auxiliary learning explanation content generation unit is used to call the auxiliary learning big model to instruct the auxiliary learning big model to combine the auxiliary learning background information and the field definition of the configured semantic unit block to generate the auxiliary learning explanation content of the target test question in the form of the semantic unit block, wherein the semantic unit block at least includes a text manuscript field, and the text manuscript field is the content of the explanation process; The voice broadcast unit is used to voice broadcast the text of the text field in the auxiliary explanation content.

18. An electronic device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement the various steps of the test question auxiliary learning and explanation method as described in any one of claims 1 to 16.

19. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the test question auxiliary learning and explanation method as described in any one of claims 1 to 16 is implemented.

20. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, each step of the test question auxiliary learning and explanation method as described in any one of claims 1 to 16 is implemented.

Images