Intelligent question answering method and device and electronic equipment

Abstract

The invention provides an intelligent question answering method and device and electronic equipment. According to the embodiment, heuristic examination and heuristic explanation are carried out based on a large model, blackboard writing information is generated while explanation is carried out in the explanation process, summarization is carried out after explanation is finished, and question recommendation is carried out according to a one-to-three way, so that the whole set of personalized teaching scheme realizes conversion from infused learning to heuristic learning, and the teaching efficiency is improved. The method has the advantages that the user is actively guided to think, the method is applied to open problem scenes, the user and the large model can be guided in a targeted manner to solve problems including but not limited to question answering, knowledge point answering and the like, and the user can be guided in a targeted manner for the thinking process when describing the thinking process.

Description

Technical Field

[0001] The present application relates to natural language processing technology, and in particular to intelligent question-answering methods, devices, and electronic equipment. Background Art

[0002] In current educational scenarios, traditional intelligent question-answering systems typically build structured knowledge bases around subject knowledge points. For example, mathematical formulas (such as the Pythagorean theorem), historical events (such as the time of the Hundred Days' Reform), or physics experimental procedures (such as measuring the acceleration of gravity) are stored in the form of standard question-answer (QA) pairs. When a user, such as a student, enters a question, the system uses keyword extraction or semantic similarity to match relevant entries in the structured knowledge base, and then returns the pre-set explanation content.

[0003] This intelligent question-answering system, leveraging a structured knowledge base, provides rapid and accurate responses to frequently asked, fixed-answer objective questions. It's well-suited for standardized test review scenarios (e.g., accessing a question bank). However, it struggles with open-ended questions and offers limited guidance. Summary of the Invention

[0004] The present application provides an intelligent question-answering method, device, and electronic device for answering open-ended questions in a guided manner, thereby avoiding indoctrination-style learning.

[0005] This application provides an intelligent question-answering method, which includes: Interacting with the user based on the question input by the user to conduct heuristic question review, guiding the user to identify the key points of the question from the input question through heuristic question review and think about the solution ideas based on the key points of the question; Interacting with the user to provide heuristic explanations of the question, analyzing each response of the user during the interaction, determining an adapted target agent module from a plurality of agent modules based on the analysis results, and controlling the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanations continue; the analysis results may be a correct response, an incorrect response, a non-positive response, or a follow-up question; and different agent modules are adapted to different analysis results; During the explanation of the question, the corresponding blackboard information is generated synchronously based on the key information involved in the explanation; After the explanation of the problem is completed, the problem is summarized and analyzed, and the user's performance in the explanation process is summarized based on the interaction in which the user participated during the explanation of the problem. The user is guided to master the knowledge point information related to the problem in a way of learning by analogy and similar problems similar to the problem are recommended to the user.

[0006] An intelligent question-answering device based on a large model, comprising: A heuristic question review module, configured to interact with the user based on the question input by the user to conduct heuristic question review, and guide the user to identify the key points of the question from the input question through heuristic question review and think about the solution ideas based on the key points of the question; a guided explanation module for interacting with the user to provide heuristic explanations of the question, and analyzing each user's response during the interaction, determining an adapted target agent module from a plurality of agent modules based on the analysis results, and controlling the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanation continues; the analysis results may be a correct response, an incorrect response, a non-positive response, or a follow-up question; different agent modules are adapted for different analysis results; The blackboard writing module is used to synchronously generate corresponding blackboard writing information based on the key information involved in the explanation during the problem explanation process; The summary and reply module is used to summarize and analyze the problem after the explanation is completed, and summarize the user's performance in the explanation process based on the interaction in which the user participates during the explanation of the problem, guide the user to master the knowledge point information related to the problem in a way of learning by analogy, and recommend similar problems similar to the problem to the user.

[0007] An embodiment of the present application further provides an electronic device. The electronic device includes: a processor and a machine-readable storage medium; The machine-readable storage medium stores machine-executable instructions that can be executed by the processor; The processor is used to execute machine-executable instructions to implement the steps of the above-disclosed method.

[0008] It can be seen from the above technical solutions that this embodiment uses the big model to conduct heuristic question review and heuristic explanation, generates blackboard information while explaining, and summarizes and recommends questions in a way of learning by analogy after the explanation. This whole set of personalized teaching solutions realizes the transformation from indoctrination learning to heuristic learning, plays a role in actively guiding users to think, and is applied to scenarios of open-ended problems. It can also guide users to interact with the big model to discuss how to solve problems, including but not limited to answering questions and knowledge points, and can provide targeted guidance on the thinking process when the user describes the thinking process.

[0009] Furthermore, in this embodiment, the Socratic guided teaching method is used to split the explanation content into multiple steps, guiding the user to complete them step by step, and analyzing each user's response during the guidance process. Based on the analysis results, an adapted target agent module is determined from multiple agent modules to control the target agent module to perform operations adapted to the analysis results based on the analysis results, so that the heuristic explanation continues to be executed. This adaptive execution of the corresponding agent module based on the user's response can give the user an appropriate response in a timely manner and improve the interactive perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0011] Figure 1 A flow chart of the method provided in the embodiment of the present application; Figure 2 Recommended schematic diagram provided for the embodiment of this application; Figure 3 Flowchart for implementing step 101 provided in the embodiment of the present application; Figure 4 Flowchart for implementing step 102 provided in the embodiment of the present application; Figure 5 Provided in the embodiments of this application Figure 4 Example application diagram; Figure 6 A diagram of the device structure provided in an embodiment of the present application; Figure 7 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The embodiment of the present application provides a method for intelligently answering questions in a multimodal manner using a large model. The method can guide users to answer questions in a Socratic way, including question review, discussion and explanation, blackboard writing, and drawing inferences from one example. It can teach students in accordance with their aptitude and provide users with flexible and personalized teaching plans. In order to enable those skilled in the art to better understand the technical solutions provided by the embodiment of the present application and to make the above-mentioned purposes, features and advantages of the embodiment of the present application more obvious and easy to understand, the technical solutions in the embodiment of the present application are further described in detail below with reference to the accompanying drawings.

[0013] See also Figure 1 , Figure 1 This is a flowchart of a method provided in an embodiment of the present application. This process is applied to electronic devices such as user terminals. The user terminals here are devices such as computers, learning machines, etc., and this embodiment does not specifically limit them.

[0014] like Figure 1 As shown, the process may include the following steps: Step 101 , based on the question input by the user, the system interacts with the user to conduct heuristic question review, and guides the user to identify the key points of the question from the input question through the heuristic question review and think about the solution ideas based on the key points of the question.

[0015] This step 101 can be implemented using a large model. This large model is based on an existing open-source large model and undergoes supervised fine-tuning (SFT) to obtain a task model for question answering. Specifically, a heuristic question review module can be set up within the large model to implement step 101.

[0016] Furthermore, in this embodiment, the format of the input question can be a picture format, a voice format, a text format, or a mixed picture and text format, etc., which is not specifically limited in this embodiment.

[0017] In this embodiment, when a user, such as a student, enters a question into the large model, the large model interacts with the user based on the question input to perform a heuristic review. This heuristic review guides the user to identify the key points in the question input and to think about the solution based on these key points. This approach, rather than directly instilling answers to questions, encourages the user to actively think about the solution and teaches the user problem-solving methods. The following examples describe how this heuristic review works and will not be elaborated on here.

[0018] Step 102: interact with the user to provide a heuristic explanation of the problem, and analyze each response of the user during the interaction process, determine an adapted target agent module from multiple agent modules based on the analysis results, and control the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanation continues to be executed; and during the problem explanation process, synchronously generate corresponding blackboard information based on the key information involved in the explanation.

[0019] This step 102 can be implemented with the aid of the aforementioned large model. For example, a heuristic question-examination module can be provided in the aforementioned large model, and step 102 can be implemented through the heuristic question-examination module.

[0020] In this embodiment, the large model does not directly output the answer to the question, but interacts with the user to provide heuristic explanations of the problem, actively guides the user to think, guides the user to discuss how to solve the problem, and analyzes each user's reply during the guidance process, such as analyzing whether the reply is correct, analyzing whether the reply is not a positive reply, or analyzing whether there are follow-up questions in the reply, etc. Then, based on the analysis results, an adapted target agent module is determined from multiple agent modules to control the target agent module to perform operations adapted to the analysis results based on the analysis results, so that the heuristic explanation continues to be executed, realizing the transition from indoctrination learning to heuristic learning, and achieving the purpose of allowing users to think actively and teaching users how to solve problems.

[0021] As for how to determine the adapted target proxy module from multiple proxy modules based on the analysis results, an example will be given below and will not be repeated here.

[0022] It should be noted that, in this embodiment, the above-mentioned multiple agent modules can be configured in the above-mentioned large model or can be independent of the large model, which is not specifically limited in this embodiment.

[0023] Furthermore, as described in step 102, during the explanation process, this embodiment will synchronously generate corresponding blackboard information based on the key information involved in the explanation. For example, like a real teacher, the key information involved in the explanation is written on the blackboard. This can not only provide users with a real-life explanation experience, but also help users organize and understand the explanation process. It should be noted that in order to facilitate the generation of blackboard information, this embodiment will also perform supervised fine-tuning on the above-mentioned large model to synchronously generate corresponding blackboard information as the explanation progresses.

[0024] In this embodiment, the above-mentioned blackboard information includes the key information involved in the explanation. Generally speaking, the refinement requirements of the blackboard information are higher than the refinement requirements of the explanation. That is, the blackboard information is usually more refined, while the explanation content is relatively more detailed. Therefore, when generating the blackboard content, the explanation content cannot be directly reused. It is necessary to generate a more refined blackboard while explaining during the explanation process. Compared with the explanation content, the blackboard information includes, for example: key problem-solving steps, methods, calculation formulas, etc. Table 1 shows examples of questions, explanation content, and blackboard information: Table 1

[0025] In this embodiment, the blackboard information is generated as the explanation progresses, and finally the explanation + blackboard information are output synchronously. It should be noted that the explanation here can be in the form of voice or text, etc., which is not specifically limited in this embodiment.

[0026] Step 103, after the explanation of the problem is completed, the problem is summarized and analyzed, and the user's performance in the explanation process is summarized based on the interaction in which the user participates during the explanation of the problem, and the user is guided to master the knowledge points related to the problem in a way of learning by analogy and recommending similar problems to the user.

[0027] This step 103 can be implemented with the help of the above-mentioned large model. For example, a summary reply agent module is set in the above-mentioned large model, and step 103 is implemented through the summary reply agent module.

[0028] In this embodiment, after the explanation is completed, the questions are summarized and analyzed. The user's performance during the explanation, such as cooperation, distraction, and user interaction behaviors, is also summarized. Finally, similar questions can be recommended to the user based on the principle of learning from one example. This allows users to master the solution to a class of problems based on their understanding of the current problem, thereby improving their learning outcomes.

[0029] Optionally, in this embodiment, the questions are summarized and analyzed, such as summarizing the parts that the user has mastered well during the explanation process, such as the parts that the user answered correctly, the corresponding knowledge points and other information, and summarizing the parts that the user answered incorrectly during the explanation process, such as the parts that reflect unfamiliarity with certain knowledge points or certain types of problem-solving ideas; and / or, guiding the user to review the knowledge points tested by the current question, and / or, asking the user questions about the knowledge points or key parts of the knowledge points, and / or, guiding the user to summarize the type of questions and applicable problem-solving methods.

[0030] In addition, in this embodiment, there are many ways to recommend similar questions to the user. For example, search the question bank for questions related to the knowledge points examined in the current question, and / or search the question bank for questions that meet the text similarity requirements with the current question; the requirements here include text similarity being greater than a set value. For another example, first obtain the common error-prone points of multiple users on the type of question to which the question belongs that have been mined; rewrite the question based on the common error-prone points and the personalized error-prone points of the user's current explanation of the question, and recommend the rewritten question. The rewritten question involves the common error-prone points and the personalized error-prone points. Here, the common error-prone points are extracted based on the user's (taking students as an example) answer record library and teaching plan library, as shown below. Figure 2 shown.

[0031] So far, completed Figure 1 The process shown.

[0032] pass Figure 1As can be seen from the shown process, this embodiment uses the big model to conduct heuristic question review and heuristic explanation, generates blackboard information while explaining, and summarizes and recommends questions in a way of learning by analogy after the explanation. This whole set of personalized teaching plans realizes the transformation from indoctrination learning to heuristic learning, plays a role in actively guiding users to think, and is applied to scenarios of open-ended problems. It can also guide users to interact with the big model to discuss how to solve problems, including but not limited to answering questions and knowledge points.

[0033] Furthermore, in this embodiment, the Socratic guided teaching method is used to split the explanation content into multiple steps, guiding the user to complete them step by step, and analyzing each user's response during the guidance process. Based on the analysis results, an adapted target agent module is determined from multiple agent modules to control the target agent module to perform operations adapted to the analysis results based on the analysis results, so that the heuristic explanation continues to be executed. This adaptive execution of the corresponding agent module based on the user's response can give the user an appropriate response in a timely manner and improve the interactive perception.

[0034] The following describes the interaction with the user based on the question input by the user in step 101 to perform heuristic question review: See also Figure 3 , Figure 3 This is a flow chart of the method provided in the embodiment of this application. Figure 3 As shown, the process may include: Step 301: Segment the input question.

[0035] For example, the input question is segmented into sentences through the above-mentioned large model (specifically, the heuristic question review module).

[0036] Step 302: for each sentence, identify the sentence type of the sentence, extract the key question in the sentence based on the sentence type, and ask guided questions based on the key question to trigger the user to reply based on the guided questions.

[0037] In this embodiment, different sentence types, such as declarative sentences, interrogative sentences, etc., contain different key information. For ease of understanding, Table 2 shows an example of the question review process: Table 2

[0038] It should be noted that, in this embodiment, during the heuristic question review process, each reply of the user is analyzed; if it is analyzed that there is a need for follow-up questions in the reply, the follow-up question reply agent module is called to analyze the reasons why the user initiated the follow-up questions through the follow-up question reply agent module, and provide personalized answers based on the reasons. When it is confirmed that the user no longer has any questions about the follow-up questions, the heuristic question review is triggered to continue until the review is completed; if it is analyzed that there are errors in the reply, the knowledge point reply agent module is called to identify the knowledge point that caused the error, guide the user to learn and understand the knowledge point, and trigger the heuristic question review to continue after confirming that the user understands the knowledge point, until the review is completed; if it is analyzed that the reply content is correct, the heuristic question review is continued through interaction with the user until the review is completed.

[0039] So far, completed Figure 3 The process shown.

[0040] pass Figure 3 The process shown implements how to interact with the user to perform heuristic question review based on the questions input by the user.

[0041] The following describes how to determine the adapted target proxy module from multiple proxy modules based on the analysis result in step 102 by way of example: See also Figure 4 , Figure 4 The step 102 provided in the embodiment of the present application is a flowchart. Figure 4 As shown, the process may include the following steps: Step 401: If the analysis result shows that the reply is correct, the target agent module is determined to be the explanation reply agent module, so that the explanation process is advanced by the explanation reply agent module.

[0042] Optionally, if the analysis result shows that the reply is correct, praise information can be further output to praise the user and continue the explanation.

[0043] In step 402, if the analysis result indicates that there is a follow-up question in the reply, the target agent module is determined to be the follow-up question reply agent module, so as to analyze the reason why the user initiated the follow-up question through the follow-up question reply agent module, and provide personalized answers based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the explanation reply agent module is triggered to continue the explanation process.

[0044] That is, when there are follow-up questions in the user's reply, it is necessary to analyze the reason for the follow-up question and provide personalized answers based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the explanation reply agent module is triggered to continue the explanation process and return to the original explanation content. For example, the following table 3 shows an example of a situation where there are follow-up questions in the reply: Table 3

[0045] In Table 3, the student’s response “Why? M-1=2, n+1=2” indicates that there is a follow-up question in the response.

[0046] In step 403, if the analysis result indicates that there is an error in the reply, the target agent module is determined to be the knowledge point reply agent module, so as to identify the knowledge point that causes the error through the knowledge point reply agent module, guide the user to learn and understand the knowledge point, and trigger the explanation reply agent module to continue the explanation process after confirming that the user understands the knowledge point.

[0047] That is, if the analysis results indicate that there are errors in the reply, the cause of the error in the user's reply will be diagnosed. First, the user's weak knowledge points will be explained. During this period, corresponding learning materials can be recommended, or simple questions for a single knowledge point can be recommended to guide the user to learn the knowledge point, and then the correct explanation will be told to the user; after confirming that the user understands the knowledge point, the explanation process will be continued by triggering the explanation reply agent module to return to the original explanation content.

[0048] It should be noted that, in this embodiment, the user is guided to learn and understand the knowledge point, and the knowledge point can be output through video and / or text guidance. The video here can be a video found in the knowledge base according to the knowledge point.

[0049] It should also be noted that the analysis result here indicates that there are errors in the reply, such as an incomplete reply, a partial or complete reply error, etc., which is not specifically limited in this embodiment. Table 4 shows an example of a situation where the analysis result indicates that there are errors in the reply: Table 4

[0050] Step 404 : If the analysis result indicates that the user has not responded positively, the target agent module is determined to be the explanation and reply agent module, so that the explanation and reply agent module guides the user to the original topic to advance the explanation process.

[0051] The analysis result here indicates that the user did not respond directly, for example, the user responded with irrelevant content, such as: asfr, the weather is really nice today; or the user only advanced the explanation, such as: trigger to continue, etc.

[0052] So far, completed Figure 4 The process shown.

[0053] To make Figure 4 The process shown is clearer. Figure 5 An example shows Figure 4 Example structure diagram of the process shown.

[0054] The above is an analysis of the method provided in the embodiment of the present application. The following describes the device provided in the embodiment of the present application: See also Figure 6 , Figure 6 This is a diagram of the device structure provided in an embodiment of the present application. The device may include: A heuristic question review module, configured to interact with the user based on the question input by the user to conduct heuristic question review, and guide the user to identify the key points of the question from the input question through heuristic question review and think about the solution ideas based on the key points of the question; a guided explanation module for interacting with the user to provide heuristic explanations of the question, and analyzing each user's response during the interaction, determining an adapted target agent module from a plurality of agent modules based on the analysis results, and controlling the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanation continues; the analysis results may be a correct response, an incorrect response, a non-positive response, or a follow-up question; different agent modules are adapted for different analysis results; The blackboard writing module is used to synchronously generate corresponding blackboard writing information based on the key information involved in the explanation during the problem explanation process; The summary and reply module is used to summarize and analyze the problem after the explanation is completed, and summarize the user's performance in the explanation process based on the interaction in which the user participates during the explanation of the problem, guide the user to master the knowledge point information related to the problem in a way of learning by analogy, and recommend similar problems similar to the problem to the user.

[0055] The heuristic question review module further analyzes each reply of the user during the heuristic question review process; if it is analyzed that there is a need for follow-up questions in the reply, the follow-up question reply agent module is called to analyze the reason why the user initiated the follow-up question through the follow-up question reply agent module, and provide personalized answers based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the heuristic question review is triggered to continue until the review is completed; if it is analyzed that there is an error in the reply, the knowledge point reply agent module is called to identify the knowledge point that caused the error, guide the user to learn and understand the knowledge point, and trigger the heuristic question review to continue after confirming that the user understands the knowledge point, until the review is completed; if it is analyzed that the reply content is correct, the large model question review module continues to interact with the user based on the questions input by the user to conduct heuristic question review until the review is completed.

[0056] The guided explanation module determines an adapted target proxy module from the plurality of proxy modules based on the analysis result, and controls the target proxy module to perform operations adapted to the analysis result based on the analysis result, including: If the analysis result shows that the reply is correct, the target agent module is determined to be the explanation reply agent module, so that the explanation process is advanced by the explanation reply agent module; If the analysis result indicates that there is a follow-up question in the reply, the target agent module is determined to be the follow-up question reply agent module, so that the follow-up question reply agent module analyzes the reason for the user to initiate the follow-up question, and provides a personalized answer based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that there is an error in the reply, the target agent module is determined to be a knowledge point reply agent module, so that the knowledge point causing the error is identified by the knowledge point reply agent module, the user is guided to learn and understand the knowledge point, and after confirming that the user understands the knowledge point, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that the user has not responded positively, the target agent module is determined to be an explanation and reply agent module, so that the explanation and reply agent module guides the user to the original topic to advance the explanation process.

[0057] The blackboard writing module is used to synchronously generate corresponding blackboard writing information based on the key information involved in the explanation, including: performing supervised fine-tuning on the large model to synchronously generate corresponding blackboard writing information following the explanation; wherein the blackboard writing information includes the key information involved in the explanation; the refinement requirement of the blackboard writing information is higher than the refinement requirement of the explanation; and / or, After synchronously generating corresponding blackboard information based on key information involved in the explanation, the method further includes: outputting the explanation and the blackboard information; wherein the explanation is given in voice or text form.

[0058] The summary reply module recommends similar questions to the user, including: Obtaining the common error points found by multiple users on the problem type to which the problem belongs; The question is rewritten based on the common error-prone points and the personalized error-prone points made by the user in explaining the question, the rewritten question involving the common error-prone points and the personalized error-prone points; and the rewritten question is recommended.

[0059] The heuristic question review module is used to interact with the user based on the questions input by the user to perform heuristic question review, including: dividing the input questions into sentences through a large model; for each sentence, identifying the sentence type of the sentence, extracting the key questions in the sentence based on the sentence type, and asking guided questions based on the key questions to trigger the user to respond based on the guided questions.

[0060] The present application also provides Figure 6 The hardware structure of the device shown. Figure 7 , Figure 7 This is a structural diagram of an electronic device provided in an embodiment of the present application. Figure 7As shown, the hardware structure may include: a processor and a machine-readable storage medium, the machine-readable storage medium storing machine-executable instructions that can be executed by the processor; the processor is used to execute the machine-executable instructions to implement the method disclosed in the above example of this application.

[0061] Based on the same application concept as the above method, an embodiment of the present application also provides a machine-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed by a processor, the method disclosed in the above example of the present application can be implemented.

[0062] Exemplarily, the machine-readable storage medium may be any electronic, magnetic, optical, or other physical storage device that may contain or store information, such as executable instructions, data, and the like. For example, the machine-readable storage medium may be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, a storage drive (such as a hard disk drive), a solid-state drive, any type of storage disk (such as a CD, DVD, etc.), or similar storage media, or a combination thereof.

[0063] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. An intelligent question answering method, characterized in that: The method includes: Interacting with the user based on the question input by the user to conduct heuristic question review, guiding the user to identify the key points of the question from the input question through heuristic question review and think about the solution ideas based on the key points of the question; Interacting with the user to provide heuristic explanations of the question, analyzing each response of the user during the interaction, determining an adapted target agent module from a plurality of agent modules based on the analysis results, and controlling the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanations continue; the analysis results may be a correct response, an incorrect response, a non-positive response, or a follow-up question; and different agent modules are adapted to different analysis results; During the explanation of the question, the corresponding blackboard information is generated synchronously based on the key information involved in the explanation; After the explanation of the problem is completed, the problem is summarized and analyzed, and the user's performance in the explanation process is summarized based on the interaction in which the user participated during the explanation of the problem. The user is guided to master the knowledge point information related to the problem in a way of learning by analogy and similar problems similar to the problem are recommended to the user.

2. The method according to claim 1, characterized in that The method further comprises: During the heuristic review process, each response of the user is analyzed; If the analysis shows that the reply requires a follow-up question, the follow-up question reply agent module is called to analyze the reason for the user to initiate the follow-up question through the follow-up question reply agent module, and provide personalized answers based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the heuristic question review is triggered to continue until the question review is completed; If the reply is found to have an error, the knowledge point reply agent module is called to identify the knowledge point that caused the error, guide the user to learn and understand the knowledge point, and trigger the heuristic question review to continue after confirming that the user understands the knowledge point until the question review is completed; If the response is analyzed to be correct, continue to interact with the user to conduct heuristic question review until the review is completed.

3. The method according to claim 1, characterized in that The step of determining an adapted target proxy module from a plurality of proxy modules based on the analysis result, and controlling the target proxy module to perform an operation adapted to the analysis result based on the analysis result includes: If the analysis result shows that the reply is correct, the target agent module is determined to be the explanation reply agent module, so that the explanation process is advanced by the explanation reply agent module; If the analysis result indicates that there is a follow-up question in the reply, the target agent module is determined to be the follow-up question reply agent module, so that the follow-up question reply agent module analyzes the reason for the user to initiate the follow-up question, and provides a personalized answer based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that there is an error in the reply, the target agent module is determined to be a knowledge point reply agent module, so that the knowledge point causing the error is identified by the knowledge point reply agent module, the user is guided to learn and understand the knowledge point, and after confirming that the user understands the knowledge point, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that the user has not responded positively, the target agent module is determined to be an explanation and reply agent module, so that the explanation and reply agent module guides the user to the original topic to advance the explanation process.

4. The method according to claim 1, wherein The synchronous generation of corresponding blackboard information based on the key information involved in the explanation includes: By performing supervised fine-tuning on the large model, corresponding blackboard information is generated synchronously with the explanation; wherein, the blackboard information includes key information involved in the explanation; and the refinement requirement of the blackboard information is higher than the refinement requirement of the explanation.

5. The method according to claim 1 or 4, characterized in that After the corresponding blackboard information is synchronously generated based on the key information involved in the explanation, the following is further included: Output explanation and blackboard information; The explanation is given in the form of voice; or the explanation is given in the form of text.

6. The method according to claim 1, characterized in that The recommending to the user similar questions as the question includes: Obtaining the common error points found by multiple users on the problem type to which the problem belongs; Rewriting the question based on the common error-prone points and the personalized error-prone points of the user during the explanation of the question, wherein the rewritten question involves the common error-prone points and the personalized error-prone points; Recommend a rewritten question.

7. The method according to claim 1, characterized in that The interacting with the user based on the question input by the user to perform heuristic question review includes: Sentence the input question; For each sentence, the sentence type of the sentence is identified, the key question in the sentence is extracted based on the sentence type, and a guided question is asked based on the key question to trigger the user to reply based on the guided question.

8. An intelligent question-answering device, characterized in that: The device includes: A heuristic question review module, configured to interact with the user based on the question input by the user to conduct heuristic question review, and guide the user to identify the key points of the question from the input question through heuristic question review and think about the solution ideas based on the key points of the question; a guided explanation module for interacting with the user to provide heuristic explanations of the question, and analyzing each user's response during the interaction process, determining an adapted target agent module from a plurality of agent modules based on the analysis results, and controlling the target agent module to perform an operation adapted to the analysis results based on the analysis results, so that the heuristic explanation continues; the analysis results may be correct responses, incorrect responses, non-positive responses, or follow-up questions; different agent modules are adapted for different analysis results; The blackboard writing module is used to synchronously generate corresponding blackboard writing information based on the key information involved in the explanation during the problem explanation process; The summary and reply module is used to summarize and analyze the problem after the explanation is completed, and summarize the user's performance in the explanation process based on the interaction in which the user participates during the explanation of the problem, guide the user to master the knowledge point information related to the problem in a way of learning by analogy, and recommend similar problems similar to the problem to the user.

9. The device according to claim 8, characterized in that The heuristic question review module further analyzes each reply of the user during the heuristic question review process; if it is analyzed that there is a need for follow-up questions in the reply, the follow-up question reply agent module is called to analyze the reason why the user initiated the follow-up question through the follow-up question reply agent module, and provide personalized answers based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the heuristic question review is triggered to continue until the review is completed; if it is analyzed that there is an error in the reply, the knowledge point reply agent module is called to identify the knowledge point that caused the error, guide the user to learn and understand the knowledge point, and trigger the heuristic question review to continue after confirming that the user understands the knowledge point, until the review is completed; if it is analyzed that the reply content is correct, the large model question review module continues to interact with the user based on the question input by the user to conduct heuristic question review until the review is completed; and / or, The step of determining an adapted target proxy module from a plurality of proxy modules based on the analysis result, and controlling the target proxy module to perform an operation adapted to the analysis result based on the analysis result includes: If the analysis result shows that the reply is correct, the target agent module is determined to be the explanation reply agent module, so that the explanation process is advanced by the explanation reply agent module; If the analysis result indicates that there is a follow-up question in the reply, the target agent module is determined to be the follow-up question reply agent module, so that the follow-up question reply agent module analyzes the reason for the user to initiate the follow-up question, and provides a personalized answer based on the reason. When it is confirmed that the user no longer has any questions about the follow-up question, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that there is an error in the reply, the target agent module is determined to be a knowledge point reply agent module, so that the knowledge point causing the error is identified by the knowledge point reply agent module, the user is guided to learn and understand the knowledge point, and after confirming that the user understands the knowledge point, the explanation reply agent module is triggered to continue the explanation process; If the analysis result indicates that the user has not responded positively, determining the target agent module as the explanation and reply agent module, so that the explanation and reply agent module guides the user to the original topic to advance the explanation process; and / or, The synchronous generation of corresponding blackboard information based on the key information involved in the explanation includes: fine-tuning the large model in a supervised manner to synchronously generate corresponding blackboard information following the explanation; wherein the blackboard information includes the key information involved in the explanation; the refinement requirement of the blackboard information is higher than the refinement requirement of the explanation; and / or, After synchronously generating corresponding blackboard information based on key information involved in the explanation, the method further includes: outputting the explanation and the blackboard information; wherein the explanation is in the form of voice or text; and / or, The recommending to the user similar questions as the question includes: Obtaining the common error points found by multiple users on the problem type to which the problem belongs; Rewrite the question based on the common error points and the user's personalized error points during the explanation of the question, where the rewritten question involves the common error points and the personalized error points; recommend the rewritten question; and / or, The interaction with the user based on the question input by the user to perform heuristic question review includes: segmenting the input question into sentences through a large model; for each sentence, identifying the sentence type of the sentence, extracting the key question in the sentence based on the sentence type, and asking guided questions based on the key question to trigger the user to reply based on the guided question.

10. An electronic device, characterized in that: The electronic device includes: a processor and a machine-readable storage medium; The machine-readable storage medium stores machine-executable instructions that can be executed by the processor; The processor is configured to execute machine-executable instructions to implement the method steps of any one of claims 1-7.

Images