Artificial intelligence-based error cause identification method and device, equipment and medium
Through the draft paper files uploaded by the user, the learners' problem-solving ideas and steps are used to analyze the learners' problem-solving ideas and steps, and accurately identify the wrong causes, solving the problem that existing intelligent learning machines cannot accurately identify the wrong causes, realizing personalized guidance, and improving Learning results.
Patent Information
- Application Number
- CN202411134042.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-29
AI Technical Summary
Existing intelligent learning machines are difficult to accurately identify the learners’ wrong reasons during the answering process, resulting in the inability to provide personalized guidance, which affects the learning effect.
Through the draft paper files uploaded by the user, the training error analysis model uses the problem-solving ideas and steps, accurately locate the error factors, display them on the answer page, and provide personalized guidance.
It improves the accuracy of identifying error causes, and can provide personalized guidance based on learners' error causes, enhance learning motivation, and narrow the gap in educational resource allocation.
Smart Images

Figure CN119047564A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent education and artificial intelligence technology, and in particular to an error cause identification method, device, equipment and medium based on artificial intelligence. Background Art
[0002] With the development of artificial intelligence technology, it is widely used in the field of education, and there are various teaching products based on artificial intelligence, such as step-by-step analysis products for test questions, test paper correction products, etc. Moreover, the education model is gradually shifting from the traditional teacher-centered to the student-centered one, and the existing system urgently needs to adapt to this change.
[0003] Generally, the existing technology will detect the pause time between the last answer and the current answer and the student's sitting posture during the user's answering process to determine whether the user has encountered difficulty in answering the question, and then give a prompt for the question. However, this method may not accurately distinguish whether the user has really stopped answering the question or is just thinking, resulting in unnecessary interruptions and irrelevant prompts, thus affecting the student's thinking.
[0004] Of course, after answering the questions, the students can simply judge whether the answers are right or wrong based on the answers they provide. However, this binary result cannot explain in depth whether the students made mistakes due to misunderstanding of the concepts, errors in the calculation process, or the fact that they simply did not master the knowledge points related to the question. In addition, due to individual differences in the mastery and understanding of knowledge among learners, existing intelligent learning machines often cannot provide customized guidance and suggestions based on the individual needs of learners, lack interaction with learners on questions during answering, and are difficult to effectively stimulate learners' thinking.
[0005] Therefore, how to personalize learning for learners is crucial. Summary of the invention
[0006] The present invention provides an artificial intelligence-based error cause identification method, device, equipment and medium to improve the accuracy of error cause identification.
[0007] According to one aspect of the present invention, there is provided a method for error cause identification based on artificial intelligence, the method comprising:
[0008] In response to an upload operation of at least one draft paper file for at least one target question by a user, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question;
[0009] Obtain at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps;
[0010] According to the current user error cause, the error cause analysis result of the draft paper file is determined, and the error cause analysis result is displayed on the answer page of the target question.
[0011] According to another aspect of the present invention, there is provided an error cause identification device based on artificial intelligence, the device comprising:
[0012] An operation response module, used to respond to a user's upload operation of at least one draft paper file for at least one target question, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question;
[0013] An error cause acquisition module, used to acquire at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps;
[0014] The result generation module is used to determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question.
[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:
[0016] at least one processor; and
[0017] a memory communicatively connected to the at least one processor; wherein,
[0018] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the artificial intelligence-based error cause identification method described in the embodiment of the present invention.
[0019] According to another aspect of the present invention, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the artificial intelligence-based error cause identification method described in an embodiment of the present invention when executed.
[0020] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program, and when the computer program is executed by a processor, the error cause identification method based on artificial intelligence according to an embodiment of the present invention is implemented.
[0021] The technical solution of the embodiment of the present invention is to respond to the user's upload operation of at least one draft paper file for at least one target question, wherein the draft paper file includes one or more of the problem-solving ideas or problem-solving steps generated by the user for the target question; obtain at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps; determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question. The above technical solution, with the help of a large model to analyze the error cause of the answer file, can accurately locate the error cause, and then determine the error cause analysis result of the user's error cause, which can improve the accuracy of error cause knowledge point identification, thereby providing personalized guidance to learners.
[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0024] Figure 1 is a flow chart of an error cause identification method based on artificial intelligence provided according to an embodiment of the present invention;
[0025] Figure 2 is an example diagram of a draft paper file provided according to an embodiment of the present invention;
[0026] Figure 3 This is an example diagram of an application of an error cause identification method based on artificial intelligence in mathematics according to an embodiment of the present invention;
[0027] Figure 4 is an example diagram of a draft paper file on a mathematics subject provided according to an embodiment of the present invention;
[0028] Figure 5 is a flow chart of a method for determining an error cause analysis result provided according to an embodiment of the present invention;
[0029] Figure 6 is an example diagram of another draft paper file provided according to an embodiment of the present invention;
[0030] Figure 7is a flowchart of another method for determining an error cause analysis result provided according to an embodiment of the present invention;
[0031] Figure 8 This is an example diagram of an application of an artificial intelligence-based error cause identification method in chemistry according to an embodiment of the present invention;
[0032] Fig. 9 is an example diagram of a draft paper file on the chemistry subject provided according to an embodiment of the present invention;
[0033] Fig.10 This is an example diagram of an application of an error cause identification method based on artificial intelligence in physics according to an embodiment of the present invention;
[0034] Fig.11 is an example diagram of a draft paper file in the physics subject provided according to an embodiment of the present invention;
[0035] Fig.12 This is an example diagram of an application of an error cause identification method based on artificial intelligence in the English subject according to an embodiment of the present invention;
[0036] Fig.13 is an example diagram of a draft paper file on the English subject provided according to an embodiment of the present invention;
[0037] Fig.14 is a structural schematic diagram of an error cause identification device based on artificial intelligence provided according to an embodiment of the present invention;
[0038] Fig.15 It is a structural schematic diagram of an electronic device for implementing the error cause identification method based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION
[0039] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0040] It should be noted that the terms "first", "second", "third", "fourth", "candidate", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0041] In addition, it should be noted that in the technical solution of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of relevant data such as draft paper documents involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0042] Figure 1 This is a flow chart of an error cause identification method based on artificial intelligence provided according to an embodiment of the present invention. This embodiment can be applied to how to identify the cause of an error based on the learner's answer information in an intelligent education scenario, and is particularly applicable to how to identify the cause of an error based on the draft file information of the answer. The method can be executed by an error cause identification device based on artificial intelligence, which can be implemented in the form of hardware and / or software, and can be integrated into an electronic device that carries the error cause identification function based on artificial intelligence, such as a server. Figure 1 As shown, the method includes:
[0043] S110, responding to an upload operation of at least one draft paper file for at least one target question by a user, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question.
[0044] The technical solution of the present application can be applied to scenarios where students or users use smart devices, such as learning machines, before, during, or after formally answering a target question.
[0045] In one embodiment, the same question answering interface may include two or three questions, or even different questions may belong to different subjects. The target questions may involve various subjects, such as mathematics, chemistry, physics, Chinese, etc.
[0046] In one case, if the user is in a mixed answering scenario, for example, a student is in a mixed answering mode for multiple subject questions, before determining the target grade corresponding to the target question, if necessary, it is also possible to first determine whether the subject corresponding to the target question is mathematics, physics, or geography, so as to more accurately determine the grade corresponding to the question, so as to facilitate the identification of errors for different grades.
[0047] Of course, by analogy, the draft files uploaded by students may also include generated problem-solving steps or problem-solving ideas for multiple questions.
[0048] The draft paper file includes one or more of the solution ideas or solution steps generated by the user for the math problem. For example, it can be the solution ideas written on the draft paper for the math problem, or it can be the solution steps written on the draft paper. Of course, optionally, it can also be a combination of the solution ideas and the solution steps.
[0049] The problem-solving ideas and steps mentioned in this embodiment can be explained by taking a chemistry problem as an example:
[0050] Target topic:
[0051] 150g of calcium carbonate was calcined at high temperature and cooled after a period of time. 22g of carbon dioxide gas was measured. What is the mass of calcium oxide in the remaining solid after the reaction? (Ca-40, C-12, O-16)
[0052] The solution in the draft document:
[0053] 1. Calcium carbonate generates calcium oxide and carbon dioxide, of which the carbon dioxide turns into gas and escapes;
[0054] 2. Calcium oxide is 56, carbon dioxide is 44, including 22 grams of carbon dioxide and 150 grams of calcium carbonate;
[0055] 3. Use a formula to determine the mass of calcium oxide.
[0056] or,
[0057] See also Figure 2 The problem-solving steps in the draft file shown. This embodiment is described by taking the example that the answering interface has only one question and only includes the draft paper of the question.
[0058] Among them, the target question refers to the question that needs to identify the cause of the error, such as Figure 3A math problem is given. A draft paper file refers to a file that contains all or part of the process of students or users solving the target problem, specifically refers to the draft paper generated by students or users in the process of solving one or several problems; the draft paper can be written on a carrier, such as a drawing board, calculation paper, open space, or even on the palm of the hand; students or users can upload the handwritten draft paper to the server or device by scanning, taking photos, taking videos, copying on the handwriting screen, or voice recording.
[0059] Optionally, it can also be the content written by the student on the draft paper in the draft paper function on the current answer page. When the student submits the draft paper, or completes the answer, one or more pages of the draft paper are automatically or manually selected to be submitted to the server for processing.
[0060] In another alternative, the scratch paper file may refer to audio, or video.
[0061] When users, or students, are unable to use pen and paper conveniently, they can click the recording or video recording control on the draft paper function to record their problem-solving ideas and steps through narration.
[0062] This can greatly facilitate students to record their problem-solving ideas and can be used by students to review their own problem-solving process.
[0063] Of course, in an optional implementation scheme, it is not limited to using the recording and video recording functions of the draft paper function in the learning system to collect and upload the draft paper files. Students can also collect the draft paper files through the recording and video recording functions of the learning machine and mobile phone, and upload them to the server in a certain way.
[0064] The beneficial effect of this technical solution is that if a student is not in the process of solving a problem and suddenly thinks of the key to solving the problem while walking on the road, he can collect a draft paper in this way in advance, and then upload the draft paper file in the same math problem to verify his own problem-solving ideas or solve the math problem. Through this method, a variety of draft paper collection methods can be used to timely record students' problem-solving ideas and students' inspiration for the questions, so as to accurately analyze the reasons for students' mistakes later.
[0065] Furthermore, it is optional to respond to the user's answer operation on the target question so that when the electronic draft paper is detected, the electronic draft paper is used as a draft paper file; and then respond to the user's upload operation on the electronic draft paper to upload the electronic draft paper to the server.
[0066] The answering operation refers to the operation of the user starting to answer the question, such as the operation of the user starting to write on the handwriting screen. The electronic draft paper refers to the draft in digital format of one or more of the solution ideas or steps generated by the user for the target question.
[0067] In order to make the present application clearer, in one embodiment, Figure 4 As shown, taking an intelligent learning machine with a handwriting display screen and a user writing in the draft paper area of a math problem answering page to generate a draft paper as an example, the innovation and technical implementation of this application are explained.
[0068] In addition, using electronic draft papers uploaded by students that contain their ideas and steps for answering the math problems can avoid technical problems such as unclear images and unclear content recognition caused by taking photos and uploading the contents of paper draft papers. Moreover, the draft paper files containing the answering steps contain more information about students' answers, which significantly improves the accuracy of subsequent error cause identification.
[0069] In this embodiment, by using the unique resource of draft pictures as a breakthrough, it is possible to discover the various causes of students' mistakes.
[0070] S120, obtaining at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps.
[0071] In this embodiment, the error cause analysis model can be a large language model (LLM). LLM will first encode the pictures and texts in the received draft paper file respectively, convert them into tokens that the model can understand, and then perform mathematical matrix operations. It mainly uses invisible neural networks such as transformers to infer questions and answers based on the input draft paper, derive the logical chain of the answering process of the question, summarize the error causes and output them.
[0072] Specifically, the model locates the causes of students' mistakes by comparing the contents of the draft paper with the standard answers generated by the model or obtaining the standard answers from the question bank.
[0073] This embodiment can output the answer ideas or reasons for errors in the answering steps based on the user's draft paper before the user formally answers the question by comparing the differences between the draft paper and the standard answer, thereby providing students with timely feedback and personalized suggestions based on the draft paper, thereby enhancing students' learning motivation.
[0074] In one embodiment, in order to obtain more accurate answers that account for more than 90% of all error causes, such as non-knowledge point errors caused by carelessness, sloppiness, and question checking, the user can also upload the formal answer content. This solution can be used in scenarios where the formal answer content has been submitted, and the input formal answer content will be processed as part of the draft paper file.
[0075] Optionally, LLM first compares the differences between the formal answer content and the standard answer, and then combines the differences between the steps of the student's draft paper and the steps of the formal answer content analysis, and then reverses the analysis after obtaining the differences.
[0076] Specifically, if the answer content is the same as the standard answer, compare the draft paper with the official answer to identify the differences in the solution between the two, and then locate the different positions of the solution differences in the draft paper and the answer content, so as to determine the cause of the error based on the contextual information of the difference position. The difference position refers to the position of the solution difference in the draft paper and the answer content in the solution steps, and the contextual information refers to the first and last steps of the solution steps corresponding to the difference position.
[0077] It should be noted that the current user error reason refers to the reason why the user made a wrong answer based on the draft paper file uploaded by the user. When the draft paper file only includes the draft paper, the error reason in the draft paper is obtained. When the draft paper file includes the draft paper and the formal answer content, in one case, there is no error in the solution idea or solution steps in the draft paper, but there is an error in the formal answer content uploaded by the user, the user error reason based on the formal answer obtained from the draft paper is identified.
[0078] The identified cause of the error can be specifically located in the draft paper or the formal answer content, such as the positive and negative signs are written incorrectly in a certain step, the transcription is wrong in the last step, the formula is written incorrectly in a certain step, etc., as the cause of the error.
[0079] This embodiment uses LLM to comprehensively determine the user's error causes based on detailed analysis and logical reasoning of the draft paper, standard answers, answer analysis, and knowledge point definitions. It also reversely deduces the steps that the student may have made mistakes by comparing the answers actually filled in by the student, so as to achieve accurate control and summary of the student's error causes in different triggering scenarios and according to different error cause contexts, and provide personalized guidance opinions to the students. In particular, the recognition accuracy rate for errors caused by non-knowledge point mastery problems such as copying errors, unclear examination of questions, and carelessness is 80%.
[0080] In an improved solution, if either the formal answer or the draft paper is missing, LLM will also locate the cause of the error through the student's formal answer or draft paper.
[0081] In addition, in order to solve the technical problem that various draft papers, such as draft papers containing incomplete information, omit some key steps, omit the final answer, or have some numbers missing, which affect the accuracy of error cause identification, this embodiment can also use LLM to focus on the information of student answers, complete the logic of student drafts through the analysis of incomplete drafts and student answers, restore some missing draft information to locate the cause of the error; and then compare the logical chain with the standard answer to get the cause of the error.
[0082] In order to solve the technical problem that the LLM has weak ability to summarize knowledge points and even weaker ability to summarize knowledge points directly related to error causes, this embodiment also provides artificial samples with different error cause distributions for reference to LLM, thereby significantly improving its ability to summarize knowledge points related to error causes.
[0083] This embodiment also writes the interaction record into the system as a strong signal to recommend the most needed knowledge points to students.
[0084] Specifically, the error cases generated by the model are:
[0085] For example, Figure 3 The reason for the mistake in the math problem is: the student did not understand that the problem required forming a square, and thus calculated it as a rectangle, so the area was correct but the side length was wrong.
[0086] For example, the question is: The students of Hongxing Primary School had their eyesight checked. In the morning, 4 groups of students were checked, with 128 students in each group. In the afternoon, 3 groups of students were checked, with 234 students in each group. How many more students were checked in the afternoon than in the morning?
[0087] Standard answer: 190.
[0088] User answer: 1214.
[0089] Current user error: When calculating the number of people in the afternoon, the student added the result of multiplying 234 by 3 (702 people) to the result of the number of people in the morning (512 people) instead of subtracting them.
[0090] Optionally in one embodiment, before using the model to output error causes, historical question-answering information can also be obtained, the historical question-answering information including historical user error causes, historical error cause classifications, and historical error cause analysis results corresponding to the answers of different users to different target questions in different subjects; historical information related to exercises, such as historical user error causes, historical error cause classifications, and historical error cause analysis results, are input into a large language model, and the large language model is trained in a supervised manner until the output results of the large language model meet the preset results of historical user error causes, historical error cause classifications, and historical error cause analysis results, and the trained large language model is used as an error cause analysis model.
[0091] Optionally, before the model is used to identify the cause of the error, or after the accuracy of the model decreases, the specific technical solution for fine-tuning the model can also be implemented through the following steps:
[0092] 1) Package the questions, answers, explanations, knowledge point definitions, student answers, drafts and other information into a data set, combined with the real reasons for students' mistakes identified by experts (including specific reasons and classification of reasons);
[0093] 2) Collect at least 100,000 data sets as described in step 1);
[0094] 3) Call the exclusive fine-tuning API of a large commercial model (such as GPT4) to fine-tune the model (i.e. upload data, and the background API automatically adjusts the internal weight of the LLM);
[0095] 4) If it is an open source large model (such as the Llama series, Qianwen series, etc.), the data in step 1) can be converted into a mask series matrix (tokens) using LLM's dedicated masker, further converted into an encoder, and then input into the standard fine-tuning process (pipeline) for fine-tuning (similarly, this process is a process of using new data to modify the existing weight matrix of LLM itself).
[0096] 5) Repackage the fine-tuned model and call it.
[0097] This implementation trains a large pre-trained model through various data of exercises, which improves the diversity and richness of training data, helps the model understand and process exercise information more accurately, and the model performs better on exercises of different difficulty levels and can be generalized to problems that have not appeared before. The model can also learn effective problem-solving strategies and improve the ability to solve complex problems. More importantly, because the training data includes draft paper data of various students at various learning stages, grades, ages, subjects, and questions, it can accurately identify the various reasons for different students' mistakes, and can also provide students with targeted feedback based on the knowledge points and concepts contained in the exercises to help students build a knowledge system.
[0098] In addition, after generating the causes and analysis of learning errors through the model in this embodiment, it can also recommend learning paths based on the students' own learning situations according to their learning progress and level of understanding. The personalized learning resources provided will help narrow the gap in the allocation of educational resources.
[0099] S130. Determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question.
[0100] In this embodiment, the error cause analysis result refers to the content of the current user error cause fed back to the user based on the draft paper file uploaded by the user, and is not limited to determining only the error cause analysis result of the draft paper. For example, knowledge points, the current user error cause generated by the model, and the error cause classification of the current user error cause are all user error cause analysis.
[0101] In an alternative embodiment, see Figure 5 , according to the current user error cause, determine the error cause analysis result of the draft paper file, which can be achieved through the following steps:
[0102] S131. Locate the error cause pool where the target question is located, wherein the error cause pool is a directed graph with target question, error cause classification, candidate user error cause, candidate error cause analysis and candidate error cause knowledge points as different hierarchical nodes and subordinate relationships between nodes at each hierarchical level as directed edges. The error cause pool includes different candidate user error causes under the same target question and error cause analysis of each candidate user error cause.
[0103] In one embodiment, all user error causes are classified into "clusters" to generate a sustainable iterative "error cause pool", which refers to the summary of error causes of different students for a problem, and can include errors caused by various reasons, such as carelessness, unclear question reading, weak knowledge points, and wrong answer transcription, etc. In this embodiment, three initial error causes are set for each problem. As time goes by and the project is iterated, the error cause pool gradually becomes complete.
[0104] Therefore, in order to improve computing efficiency and save resources, when locating the student's error cause, first consider whether there is a similar candidate user error cause in the existing error cause pool. If so, the error cause analysis corresponding to the candidate user error cause or the error cause knowledge point is output as the error cause analysis result. Then, if the error cause of the exercise or a similar error cause is in the error cause pool, the current user error cause will not be updated to the error cause pool, otherwise it will be updated and supplemented to the error cause pool.
[0105] Specifically, the user error causes pre-set in the error cause pool have error cause classification. The error cause pool roughly divides the error causes into major error cause categories, and the error causes are further subdivided into minor error cause categories under the major error cause categories; a detailed error description / case description is given for each minor error cause category.
[0106] Optionally, the error causes can be summarized into knowledge point error causes and non-knowledge point error causes, where the knowledge point error cause is that the reason why the user made a mistake in doing the exercise is related to the knowledge point involved in the exercise, for example, the user has a wrong understanding of the knowledge, the user has not mastered the knowledge point, etc. For example, for a certain exercise, the user has a wrong understanding of the knowledge point of "division", a wrong multiplication of fractions, a wrong arrangement statistics, a wrong multiplication distributive law, a wrong three-dimensional structure, confusion of geometric figure attributes, a wrong proportional relationship, etc.
[0107] For example, a current user error belongs to the knowledge point error under the error category in the error pool. The error subcategory corresponding to the knowledge point error is incorrect application of the area of a circle; incorrect proportional relationship, and the corresponding description may be: incorrect understanding or application of the proportional relationship.
[0108] Another example is: the reason for the error in knowledge point is confusion in the attributes of geometric figures, and the corresponding description can be: using the diameter of a circle as the radius.
[0109] Non-knowledge point errors are caused by errors that are not related to knowledge points, including user misunderstanding of questions, logical reasoning errors, thinking inertia errors, answering skills errors, handwriting transcription errors, attention and detail errors, calculation errors, etc.
[0110] For example, for a trap question whose current user error cause belongs to the error comprehension error (non-knowledge point error) under the error cause category in the error cause pool, the error description or case description of the trap question is: intentionally designed to test common misunderstandings or negligence, and careful analysis is required to avoid misunderstandings.
[0111] For another example, the major category of a user's error cause is handwriting transcription error under non-knowledge point error cause, and the corresponding minor category of error cause is numerator and denominator written in reverse, and the corresponding description may be that the student has incorrect positions of numerator and denominator when writing fractions.
[0112] The structure of the fault pool can be as follows:
[0113] Exercise 1 (a math problem)--Candidate user error reason (the student did not understand that the problem required forming a square, and thus calculated it as a rectangle, so the area was correct but the side length was wrong)--Major error reason category (wrong knowledge point)--Minor error reason category (wrong understanding of geometric figures)--Error description (the perimeter of rectangles and squares, and the determination of squares are inaccurate).
[0114] Exercise 1 (a math problem) - candidate user error reasons (area was written as perimeter, perimeter was written as area) - major error category (handwriting transcription error) - minor error category (the student thought of A but wrote B) - description of error reason (the student thought correctly but wrote incorrectly).
[0115] Another example is:
[0116] Exercise 2 (a chemistry question) - Candidate user error cause (the student made a calculation error when calculating 56:X=44:22. The correct calculation result is 28, and the student's incorrect calculation result is 112) - Major error cause category (calculation error) - Minor error cause category (pure calculation error) - Description of error cause (an error occurred during the student's calculation process).
[0117] The candidate error cause analysis is the analysis of the user error cause pre-set in the error cause pool, specifically refers to the textual description of the reason why the student made a mistake in a certain exercise. In one embodiment, the candidate error cause knowledge point refers to the knowledge point pre-set in the error cause pool and related to the target question.
[0118] Preferably, error cause knowledge points are preset for the user error causes whose error cause category is knowledge point error causes, and the error cause pool may be in the form of:
[0119] Exercise 1 (a math problem)--Candidate user error reason (the student did not understand that the problem required forming a square, and thus calculated it as a rectangle, so the area was correct but the side length was wrong)--Error reason knowledge point (Error reason knowledge point 1, Error reason knowledge point 2, Error reason knowledge point 3) / --Major error reason category (wrong knowledge point)--Minor error reason category (wrong understanding of geometric figures)--Error description (the perimeter of rectangle and square, the judgment of square is inaccurate).
[0120] Error cause knowledge points and error cause categories can be located at the same level node.
[0121] Optionally, the error cause knowledge points and the error cause categories can also be in different hierarchical nodes. This is based on the way the error cause pool is constructed. As long as the attributes on the same hierarchical nodes can be the same, it is convenient for retrieval and is not limited here.
[0122] Optionally and exemplarily, as shown in Table 1, it is a classification list of the error cause pool.
[0123] Table 1 Classification list of error pools
[0124]
[0125]
[0126] S132. If the current user error cause of the target question exists in the error cause pool, determine the error cause classification of the current user error cause; wherein the error cause classification includes knowledge point error causes and non-knowledge point error causes.
[0127] Among them, the cause of knowledge point error refers to the error caused by students' weak grasp of a specific knowledge point or concept or misunderstanding. The cause of non-knowledge point error refers to the error caused not by insufficient grasp of knowledge points, but by other factors, which can be classified into at least one of the following under the cause of error subcategory: unclear examination of the question, carelessness, and transcription error.
[0128] In this embodiment, a scenario is used as an example to illustrate the implementation of the model for error cause classification:
[0129] The implementation logic is to use the specific error cause output by LLM, find the closest k categories in the error cause classification according to the similarity, and then compare and confirm them one by one through LLM to get the final error cause classification (both the major and minor categories of error causes can be implemented in this way). Among them, the confirmation step is to first confirm the major category (define the scope of the minor category), and then confirm the minor category.
[0130] For example:
[0131] Scenario 1:
[0132] Knowledge point name: Multiplication of three-digit numbers by two-digit numbers - multiplication and addition, multiplication and subtraction application problems;
[0133] Question: The first workshop produces 108 parts per day, which is 19 parts less than the second workshop. How many parts can the second workshop produce in 29 days?
[0134] Answer: (108+19)×29=127×29=3683 (pieces) Answer: The second workshop can produce 3683 parts in 29 days. So the answer is: 3683.
[0135] Specific cause of error: The student made a calculation error when calculating 127×29. He originally wanted to use the simple algorithm of 127×30-127, but in the last step he mistakenly copied the wrong minuend 127 instead of 29.
[0136] Related LLM knowledge points or suggestions for the cause of the error: [Knowledge points that students need to learn additionally] Simple calculation of two-digit multiplication. When one of the two numbers to be multiplied is close to a ten, the multiplication can be converted into a mixed operation of multiplication and addition and subtraction. First calculate the part of the multiplication of the ten numbers, and then subtract the excess or add the missing digits.
[0137] Student draft: Figure 6 .
[0138] Knowledge point name 1: Multiplication of three-digit numbers by two-digit numbers - multiplication and addition, multiplication and subtraction application problems.
[0139] Knowledge point definition: Application problem: 1. Given a quantity, the total quantity is A, divided into B parts, each part is C, how much is left? 2. The formula listed is a two-step calculation problem of multiplication and addition or multiplication and subtraction, and the multiplication formula is a three-digit number multiplied by a two-digit number. The formula can have parentheses or not.
[0140] Major causes of error: incorrect knowledge points.
[0141] Category definition: Errors caused by insufficient understanding of the knowledge points of the current question.
[0142] Sub-category of error cause: incorrect knowledge points, sub-category definition: core knowledge points have not been mastered.
[0143] S133. When the error cause of the current user error is classified as a knowledge point error cause, at least one candidate error cause knowledge point corresponding to the current user error cause is identified in the error cause pool, and the identified candidate error cause knowledge point is used as the error cause analysis result of the draft paper file.
[0144] S134. When the error cause of the current user error is classified as a non-knowledge point error cause, a candidate error cause analysis corresponding to the current user error cause is identified in the error cause pool as the error cause analysis result of the draft paper file.
[0145] Further, after the error cause analysis result of the draft paper file is determined, the current user error cause, the error cause analysis result and the first error cause knowledge point generated by the error cause analysis model may be updated to the error cause pool.
[0146] Optionally in one embodiment, see Figure 7 If the current user error of the target question does not exist in the error pool, the error analysis result can be determined by the following steps:
[0147] S1321. If the current user error cause of the target question does not exist in the error cause pool, obtain the error cause classification of the current user error cause generated by the error cause analysis model;
[0148] The implementation of error cause classification by identifying the error cause of the current user through the error cause analysis model can be as follows:
[0149] Obtain a first error cause knowledge point that has knowledge relevance to the current user error cause and is generated based on the current user error cause through an error cause analysis model, and then use the first error cause knowledge point that has the highest knowledge relevance to the current user error cause as a core error cause knowledge point; identify whether there is draft content related to the core error cause knowledge point from the draft paper file.
[0150] If there is no draft content related to the core error cause knowledge point in the draft content, then the error cause classification of the current user error cause is confirmed to be a knowledge point error cause.
[0151] If there is draft content related to the core error cause knowledge point in the draft content, the error cause of the current user error is determined to be classified as a non-knowledge point error cause.
[0152] In this embodiment, the core error cause knowledge point refers to the error cause knowledge point that is most highly associated with the current user's error cause.
[0153] It can be understood that by comparing the error cause knowledge points with the content in the draft paper to determine the corresponding error cause classification, the accuracy of error cause identification is improved, and the identified error cause classification is more in line with the user, thereby improving the user experience.
[0154] In addition, as a preferred implementation method, only the error causes that do not exist in the error cause pool are identified and classified, which greatly reduces the model burden and improves output efficiency.
[0155] Further, after obtaining the classification of the identification error causes:
[0156] S1322. When the error cause is classified as a non-knowledge point error cause, obtain an error cause analysis result of a draft paper file generated by an error cause analysis model based on the current user error cause.
[0157] The error cause analysis result in this step may be to directly analyze the error cause of the current user.
[0158] S1323. When the error cause is classified as a knowledge point error cause, a first error cause knowledge point generated by an error cause analysis model and having knowledge relevance to the current user error cause is obtained, wherein the knowledge relevance is determined based on the current user error cause and the subordinate relationship between the target question and the question knowledge point.
[0159] For example, Exercise 1: There are two rectangles of the same size, 22 cm long and 11 cm wide, put together to form a square. What is its perimeter? cm and its area? square cm.
[0160] The current user's mistake is: the student did not understand that the question required forming a square, and thus calculated it as a rectangle. Therefore, the area was correct, but the side length was wrong.
[0161] The knowledge points that are relevant to the current user's error cause may include:
[0162] (1) Application of the area formulas of rectangles and squares
[0163] (2) Characteristics of rectangles and squares
[0164] (3) Properties of a square
[0165] (4) Perimeter of rectangle and square
[0166] (5) Determination of square
[0167] (6) Calculation formula for the area of a rectangle or square
[0168] (7) Given the perimeter and length (or width) of a rectangle, find the width (or length)
[0169] …
[0170] The above knowledge points are all knowledge points involved in Exercise 1. Based on the current user error cause, the first knowledge point error cause associated with this knowledge point can be:
[0171] (1) Application of the area formulas of rectangles and squares
[0172] (4) Perimeter of rectangle and square
[0173] (5) Determination of square
[0174] (7) Given the perimeter and length (or width) of a rectangle, find the width (or length)
[0175] Of course, in addition to outputting the first error cause knowledge point on the answer page, all the knowledge points involved can also be displayed on the answer page for students to refer to and learn.
[0176] Among them, knowledge relevance is used to describe the degree of association between the error cause knowledge point and the current user error cause. The subordinate relationship is pre-set through a large number of experiments or experience values, and is used to describe the relationship between the user error cause, the target question and the question knowledge point.
[0177] S1324. Use the candidate error cause knowledge point that matches the first error cause knowledge point in the error cause knowledge point map as the error cause analysis result of the draft paper file.
[0178] The error cause knowledge graph is a graph composed of various knowledge points, and its main logical relationships include the sequence of knowledge points, the grade attributes of knowledge points, the front-end and back-end relationships of knowledge points, etc.
[0179] In one embodiment, the specific technical implementation of obtaining similar error cause knowledge points from the error cause knowledge point map may be:
[0180] The matrix of the "embedded" implicit data space is used to perform similarity matching with the knowledge point map. Among them, embedding modeling is a technical term in the field of machine learning. The text or image information is first converted into a mask (token) that can be understood by the computer, and then the mask sequence is converted into a dense matrix (dense matrix) through mathematical matrix transformation. Finally, a matrix is used to represent the semantics or meaning of a sentence or a picture.
[0181] Specifically, the semantic mathematical matrix of the error cause knowledge points output by each model and the semantic matrix of the knowledge points in the error cause knowledge point map are calculated respectively, and then the most similar set between the two is retrieved, and the knowledge points in the set are used as the error cause analysis results output to students.
[0182] Furthermore, when screening similar knowledge points, it is necessary to perform a one-to-one judgment on the knowledge points, that is, to judge the degree of correlation between the knowledge points related to the error cause and the knowledge points in the error cause knowledge point map.
[0183] After the error cause knowledge points are summarized, they still belong to general natural language knowledge points. In this embodiment, they also need to be aligned with preset nano professional knowledge points.
[0184] Specifically, the retrieval augmented generation framework (RAG) can be used to perform necessary steps such as coarse screening, fine screening, and large-model re-screening and scoring of relevant knowledge points.
[0185] In this embodiment, some first error cause knowledge points summarized according to the current user's error causes are generated through LLM, and the number can be 1, 2, or more; then the retrieval enhanced generation idea (Retrieval-Augmented Generation, RAG) is used to retrieve similar knowledge points in all knowledge points in the error cause knowledge point map, thereby achieving accurate matching of error cause knowledge points and homework-related knowledge points, and being able to accurately locate error cause knowledge points with an accuracy rate of 80%.
[0186] Of course, a preferred method is to define the search scope of the knowledge point map before searching. For example, before searching, first determine the subject and grade corresponding to the exercise, and then search for knowledge points under the corresponding subject and grade to reduce the amount of calculation of the model and improve the calculation speed.
[0187] Finally, LLM can be used to perform a one-to-one discrimination between the selected knowledge points and the user's error causes, and select the knowledge points that truly meet the conditions as the error cause knowledge points. In this embodiment, the top five knowledge points ranked by merit are used as the selected error cause knowledge points.
[0188] Specifically, in the LLM's judgment of knowledge points and error causes, the LLM scoring and re-ranking method is adopted, so that the LLM outputs a confidence score for relevance (for example, the score range is 1 to 10, then 1 represents the least relevant, 10 represents the most relevant, and so on), rather than simply judging whether it is relevant or not. In this way, LLM actually obtains additional information, which can be used as the basis for fine sorting.
[0189] Optionally, in one embodiment, a hybrid embedding model may be introduced to improve the accuracy of knowledge point screening.
[0190] Among them, the "embedding model" is a very important and widely used dense matrix retrieval model in the retrieval field. The "keyword model" mainly relies on the statistical distribution similarity of keywords and words in the retrieval field and keywords and words in the text library to retrieve useful information. The combination of the two is called the "hybrid embedding model".
[0191] For example, you can design a weight of 50% for keyword search and 50% for dense matrix search. Then the hybrid model will comprehensively consider the outputs of the two, and finally return a weighted (1:1) result list from each of the two result lists. If the weights are set to 30% and 70%, it is equivalent to selecting 30% of the keyword search results plus 70% of the dense matrix search results. If the weights are set to 0% and 100%, it is equivalent to using only the dense matrix for search, and vice versa.
[0192] This embodiment applies the hybrid embedding model to the RAG system, which plays a very critical role in building a direct connection between exercise analysis and knowledge point definition, and makes a huge contribution to "providing high-quality data to LLM". In particular, the efficiency of searching for mathematical exercises and knowledge points is greatly improved. For example, the search keyword expansion is about 30% higher than the traditional RAG search accuracy.
[0193] The candidate error cause knowledge points in this step refer to the error cause knowledge points pre-stored in the error cause knowledge point map. The error cause knowledge point map is pre-set through a large number of experiments or experience values, and stores a structural spectrum of the target question, error cause classification, candidate user error causes, candidate error cause analysis, and the relationship between candidate error cause knowledge points. It should be noted that the number of candidate error cause knowledge points stored in the error cause knowledge point map is greater than the number of candidate error cause knowledge points stored in the error cause pool.
[0194] Optionally in one embodiment, in order to more accurately obtain the knowledge points that users may need to learn and to accurately hit the students' weak knowledge points, after obtaining a knowledge point similar to the first error cause knowledge point from the error cause knowledge point map, it can be used as the second error cause knowledge point, and then the historical answer data of the target question can be obtained, wherein the historical answer data includes a first knowledge point set that has been learned by learners who answered the target question correctly, and a second knowledge point set that has been learned by learners who answered the target question incorrectly; at least one candidate error cause knowledge point that exists in the first knowledge point set but not in the second knowledge point set is used as the third error cause knowledge point; the second error cause knowledge point and the third error cause knowledge point are used as the error cause analysis results of the draft paper file.
[0195] For example, for question P1, student S1 got it right, while student S2 got it wrong. Student S1 has learned knowledge points A, B, C, and D, while student S2 has learned knowledge points B and D. So student S2 is likely to need knowledge points A and C. Specifically, knowledge points A and C can be understood as parts of the entire knowledge point map of the cause of the error. Assume that the entire knowledge point map library is ABCDEFG…XYZ. Student 1 has learned ABCD, and student 2 has learned BD. Then the two missing ACs are likely to be what student 2 needs. The search efficiency is relatively high in this. Otherwise, if ABCD…XYZ are all searched, the efficiency is relatively low. The search method is mainly to use the direct semantic similarity between the error cause knowledge points generated by the large model and the actual error cause knowledge points to search (top N), and then use the large model to confirm (top K).
[0196] It should be noted that, in one embodiment, the construction of the error cause knowledge point map is the same as the construction idea of the above map.
[0197] This implementation interprets key information such as knowledge point maps and student learning history, and uses the direct difference between "knowledge points learned by students who got it right" and "knowledge points learned by students who got it wrong" to create a candidate pool of knowledge points related to the cause of the error, focusing on the knowledge points required to solve the problem correctly, which can make learning more targeted and improve students' learning efficiency. In addition, excluding those knowledge points that cause students to answer questions incorrectly, focusing learning resources and time on effective knowledge points, avoiding wasting time on unimportant or wrong concepts, and helping to avoid repeating others' mistakes.
[0198] In an optional solution, it is also possible to remove from the knowledge points those knowledge points that the student has already learned and whose mastery level has reached a preset value. The preset value can be determined based on the student's historical question-answering information. For example, if there are 100 exercises corresponding to the knowledge point and the student's accuracy rate in answering the questions is 90%, it can be determined that the student's mastery level of the knowledge point has reached the preset value.
[0199] In this way, you can customize your learning plan based on your personal learning situation, achieve personalized learning, and quickly identify and correct your knowledge blind spots or misunderstandings.
[0200] In another optional embodiment, the answer page includes an answer area. Accordingly, after determining the error cause analysis results of the draft paper file according to the current user error cause, and displaying the error cause analysis results on the answer page of the target question, the second error cause knowledge point and the third error cause knowledge point can also be displayed in the bottom area of the answer area; in response to the user's click operation on the second error cause knowledge point or the third error cause knowledge point in the bottom area, the knowledge analysis content corresponding to the error cause knowledge point clicked by the user is displayed on the answer page, wherein the knowledge analysis content includes one or more of video content or graphic content.
[0201] In this embodiment, the answer area refers to an area for users to answer the target questions. The knowledge analysis content refers to the relevant content that analyzes the knowledge points of the wrong causes to facilitate the user's understanding; it can be a knowledge point explanation video, a knowledge point diagram, etc.
[0202] Based on the above embodiment, after determining the error cause of the current user, it is also possible to determine the associated topics related to at least one candidate error cause knowledge point, the first error cause knowledge point, the second error cause knowledge point, and the third error cause knowledge point corresponding to the error cause of the current user, and push the associated topics. In order to better understand the present invention, based on the above embodiment, some specific examples are provided:
[0203] For example, the target topic is related to the subject of chemistry, see Figure 8 and Fig. 9 ; In response to the user Figure 8 The chemistry questions shown are uploaded Fig. 9 The problem-solving steps in the draft paper shown; the error analysis model is based on Fig. 9 The problem-solving steps in the above example generate the current user error cause: the reaction equation of magnesium and oxygen is 2Mg+O2=2MgO, and the reaction equation is written incorrectly. However, knowledge points related to the reaction equation appear in the draft paper, and the error cause of the user error can be determined to be classified as a knowledge point error cause; if the user error cause is identified in the error cause pool, the candidate error cause knowledge point corresponding to the current user error cause identified in the error cause pool can be selected as the error cause analysis result of the draft paper file, such as: the method of balancing chemical equations, balancing chemical equations and related calculations, and the oxides of magnesium and aluminum.
[0204] For example, the target topic is related to the subject of physics, see Fig.10 and Fig.11 ; User input for Fig.10 Draft paper of the target topic shown Fig.11 ; The model analyzes it and obtains a current user error cause: when the student simplifies 5 / 2:1 / 1, the calculation is wrong. 5 should be left in the numerator and 2 in the denominator. In fact, it is written in reverse. The error cause is classified as a calculation error in non-knowledge point error causes; the candidate error cause analysis corresponding to the current user error cause can be identified in the error cause pool as the error cause analysis result of the draft paper file, such as: this question is correct from the beginning to the last step. Unfortunately, 2 / 5 was calculated as 5 / 2 when calculating the result. In the future, we must prevent the loss of points caused by such low-level mistakes.
[0205] For example, the target topic is related to English subject, Fig.12 The title shown is in English. Fig.13 The corresponding problem-solving steps are shown; the model is based on Fig.13Output the current user error cause: the student did not write down the word camera completely in the notes taken during dictation, or did not know how to spell the word camera, which resulted in the inability to choose camel, which is similar in shape and pronunciation; determine the error cause of the current user error cause and classify it into a knowledge point error cause; identify at least one candidate error cause knowledge point corresponding to the current user error cause in the error cause pool as the error cause analysis result of the draft paper file, such as: voice-target phonetic symbol / / , Voice-Target Phonetic Symbol / / , speech-target phonetic symbol / m / .
[0206] Fig.14 : is a schematic diagram of the structure of an error cause identification device based on artificial intelligence provided according to an embodiment of the present invention. This embodiment can be applied to how to identify the cause of an error based on the learner's answer information in an intelligent education scenario, and is particularly applicable to how to identify the cause of an error based on the draft file information of the answer. The device can be implemented in the form of hardware and / or software, and can be integrated into an electronic device that carries the error cause identification function based on artificial intelligence, such as a server. Fig.14 As shown, the device comprises:
[0207] An operation response module 110, for responding to a user's upload operation of at least one draft paper file for at least one target question, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question;
[0208] The error cause acquisition module 120 is used to acquire at least one error cause of the current user generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps;
[0209] The result generation module 130 is used to determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question.
[0210] The technical solution of the embodiment of the present invention is to respond to the user's upload operation of at least one draft paper file for at least one target question, wherein the draft paper file includes one or more of the problem-solving ideas or problem-solving steps generated by the user for the target question; obtain at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps; determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question. The above technical solution uses a large model to analyze the error cause of the answer file, can accurately locate the error cause, and then determine the error cause analysis result of the user's error cause, can improve the accuracy of error cause knowledge point identification, and thus provide personalized guidance to learners.
[0211] Optionally, the result generation module 130 includes:
[0212] The error cause pool locating unit 131 is used to locate the error cause pool where the target topic is located, wherein the error cause pool is a directed graph composed of target topic, error cause classification, candidate user error cause, candidate error cause analysis and candidate error cause knowledge points as different hierarchical nodes, and the subordinate relationship between the hierarchical nodes as directed edges. The error cause pool includes different candidate user error causes under the same target topic and the error cause analysis of each candidate user error cause;
[0213] The error cause classification determination unit 132 is used to determine the error cause classification of the current user error cause if the error cause of the current user for the target question exists in the error cause pool; wherein the error cause classification includes knowledge point error cause and non-knowledge point error cause;
[0214] A first result determination unit 133 is used to identify at least one candidate error cause knowledge point corresponding to the current user error cause in the error cause pool when the error cause of the current user error cause is classified as a knowledge point error cause, and use the identified candidate error cause knowledge point as the error cause analysis result of the draft paper file;
[0215] The second result determination unit 134 is used to identify a candidate error cause analysis corresponding to the current user error cause in the error cause pool when the error cause of the current user error cause is classified as a non-knowledge point error cause, as the error cause analysis result of the draft paper file.
[0216] Optionally, the result generation module 130 further includes:
[0217] The error cause classification acquisition unit 135 is used to acquire the error cause classification of the current user error cause generated by the error cause analysis model if the current user error cause of the target question does not exist in the error cause pool;
[0218] The third result determination unit 136 is used to obtain the error cause analysis result of the draft paper file generated by the error cause analysis model based on the current user error cause when the error cause is classified as a non-knowledge point error cause;
[0219] The knowledge point determination unit 137 is used to obtain a first error cause knowledge point generated by the error cause analysis model and having knowledge relevance to the current user error cause when the error cause is classified as a knowledge point error cause, wherein the knowledge relevance is determined based on the current user error cause, the subordinate relationship between the target question and the question knowledge point;
[0220] The fourth result determination unit 138 is used to use the candidate error cause knowledge point that matches the first error cause knowledge point in the error cause knowledge point graph as the error cause analysis result of the draft paper file.
[0221] Optionally, the third result determination unit 136 is further specifically configured to:
[0222] Acquire a candidate error cause knowledge point that matches the first error cause knowledge point from the error cause knowledge point map as a second error cause knowledge point;
[0223] Acquire historical answer data of the target question, wherein the historical answer data includes a set of first knowledge points learned by learners who answered the target question correctly and a set of second knowledge points learned by learners who answered the target question incorrectly;
[0224] At least one candidate error cause knowledge point that exists in the first knowledge point set but does not exist in the second knowledge point set is used as a third error cause knowledge point;
[0225] The second and third error cause knowledge points are used as the error cause analysis results of the draft paper file.
[0226] Optionally, the non-knowledge point error causes include at least one of unclear examination of the question, carelessness and transcription errors.
[0227] Optionally, the third result determination unit 136 is further specifically configured to:
[0228] After obtaining the first error cause knowledge point having knowledge relevance to the current user error cause generated by the error cause analysis model based on the current user error cause, the first error cause knowledge point having the highest knowledge relevance to the current user error cause is used as the core error cause knowledge point;
[0229] If the draft content related to the core error cause knowledge point is not identified from the draft paper file, the error cause classification of the current user error is confirmed as a non-knowledge point error cause.
[0230] Optionally, the operation response module 110 is specifically configured to:
[0231] In response to the user's answering operation on the target question, when the electronic draft paper is detected, the electronic draft paper is used as a draft paper file;
[0232] In response to the user's upload operation on the electronic draft paper.
[0233] Optionally, the answering page includes an answering area; accordingly, the result generating module 130 is further specifically used for:
[0234] After displaying the error cause analysis result on the answer page of the target question, displaying the second error cause knowledge point and the third error cause knowledge point in the bottom area of the answer area;
[0235] In response to a user clicking on the second error cause knowledge point or the third error cause knowledge point in the bottom area, the knowledge analysis content corresponding to the error cause knowledge point clicked by the user is displayed on the question answering page, wherein the knowledge analysis content includes one or more of video content or graphic content.
[0236] Optionally, the result generating module 130 is further specifically configured to:
[0237] The current user error cause, error cause analysis result, and first error cause knowledge point generated by the error cause analysis model are updated to the error cause pool.
[0238] Optionally, the operation response module 110 is further specifically configured to:
[0239] In response to a user uploading at least one draft paper file for at least one target question, one or more items of a standard answer to the target question or the user's answer content to the target question are obtained and updated as part of the draft paper file.
[0240] Optionally, the error cause acquisition module 120 is specifically configured to:
[0241] When the draft paper file includes answer content and the answer result of the answer content is correct, the answer content includes formal answer steps;
[0242] If the error analysis model identifies that the user's draft solution process for the target question is wrong, the difference between the draft solution process and the formal answering steps is located; and
[0243] The difference positions of the problem-solving differences in the draft paper file are located to instruct the error cause analysis model to generate context information based on the difference positions and determine at least one current user error cause.
[0244] Optionally, the error cause analysis model may be determined by:
[0245] Obtaining historical question-solving information, wherein the historical question-solving information includes historical user error causes, historical error cause classifications, and historical error cause analysis results corresponding to the answer contents of different users for different target questions of different subjects;
[0246] The historical user error causes, historical error cause classifications, and historical error cause analysis results are input into a large language model, and the large language model is trained in a supervised manner until the output results of the large language model meet the preset results of the historical user error causes, historical error cause classifications, and historical error cause analysis results, and the trained large language model is used as an error cause analysis model.
[0247] Optionally, the result generating module 130 is further specifically configured to:
[0248] Determine the associated topics related to at least one candidate error cause knowledge point, the first error cause knowledge point, the second error cause knowledge point, and the third error cause knowledge point corresponding to the current user error cause respectively;
[0249] Push related topics.
[0250] The artificial intelligence-based error cause identification device provided in the embodiment of the present invention can execute the artificial intelligence-based error cause identification method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.
[0251] Fig.15 is a schematic diagram of the structure of an electronic device for implementing the error cause identification method based on artificial intelligence according to an embodiment of the present invention, Fig.15 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.
[0252] like Fig.15 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0253] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0254] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as an error cause identification method based on artificial intelligence.
[0255] In some embodiments, the error cause identification method based on artificial intelligence can be implemented as a computer program, which is tangibly contained in a non-transitory computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the error cause identification method based on artificial intelligence described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the error cause identification method based on artificial intelligence in any other appropriate manner (for example, by means of firmware).
[0256] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs, which may be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general programmable processor, which may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0257] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0258] In the context of the present invention, a non-transitory computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or equipment. A non-transitory computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. Alternatively, a non-transitory computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0259] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).
[0260] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.
[0261] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.
[0262] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.
[0263] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for identifying error causes based on artificial intelligence, characterized in that: include: In response to an upload operation of at least one draft paper file for at least one target question by a user, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question; Obtain at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps; According to the current user error cause, the error cause analysis result of the draft paper file is determined, and the error cause analysis result is displayed on the answer page of the target question.
2. The method according to claim 1, characterized in that The step of determining the error cause analysis result of the draft paper file according to the current user error cause includes: Locate the error cause pool where the target topic is located, wherein the error cause pool is a directed graph with target topic, error cause classification, candidate user error causes, candidate error cause analysis and candidate error cause knowledge points as different hierarchical nodes, and subordinate relationships between nodes at each level as directed edges, and the error cause pool includes different candidate user error causes under the same target topic and error cause analysis of each candidate user error cause; If the error cause pool contains the current user error cause of the target question, determining the error cause classification of the current user error cause; wherein the error cause classification includes knowledge point error causes and non-knowledge point error causes; When the error cause of the current user error cause is classified as the knowledge point error cause, identifying at least one candidate error cause knowledge point corresponding to the current user error cause in the error cause pool, and using the identified candidate error cause knowledge point as the error cause analysis result of the draft paper file; When the error cause of the current user error is classified as the non-knowledge point error cause, a candidate error cause analysis corresponding to the current user error cause is identified in the error cause pool as the error cause analysis result of the draft paper file.
3. The method according to claim 2, characterized in that The step of determining the error cause analysis result of the draft paper file according to the current user error cause also includes: If the current user error cause of the target question does not exist in the error cause pool, obtaining the error cause classification of the current user error cause generated by the error cause analysis model; When the error cause is classified as a non-knowledge point error cause, obtaining an error cause analysis result of the draft paper file generated by the error cause analysis model based on the current user error cause; When the error cause is classified as a knowledge point error cause, obtaining a first error cause knowledge point generated by the error cause analysis model and having knowledge relevance to the current user error cause, wherein the knowledge relevance is determined based on the current user error cause and the subordinate relationship between the target question and the question knowledge point; The candidate error cause knowledge point in the error cause knowledge point map that matches the first error cause knowledge point is used as the error cause analysis result of the draft paper file.
4. The method according to claim 3, characterized in that The step of using the candidate error cause knowledge point in the error cause knowledge point graph that matches the first error cause knowledge point as the error cause parsing result of the draft paper file includes: Acquire a candidate error cause knowledge point that matches the first error cause knowledge point from the error cause knowledge point map as a second error cause knowledge point; Acquire historical answer data of the target question, wherein the historical answer data includes a first set of knowledge points learned by learners who answered the target question correctly and a second set of knowledge points learned by learners who answered the target question incorrectly; At least one candidate error cause knowledge point that exists in the first knowledge point set but does not exist in the second knowledge point set is used as a third error cause knowledge point; The second error cause knowledge point and the third error cause knowledge point are used as the error cause analysis result of the draft paper file.
5. The method according to claim 2, characterized in that: The non-knowledge point error causes include at least one of unclear examination of the question, carelessness and transcription errors.
6. The method according to claim 1, characterized in that The step of responding to the user uploading at least one draft paper file for at least one target topic includes: In response to the user's answering operation on the target question, when the electronic draft paper is detected, the electronic draft paper is used as the draft paper file; In response to a user's upload operation on the electronic draft paper.
7. The method according to claim 4, characterized in that The answer page includes an answer area; accordingly, after determining the error cause analysis result of the draft paper file according to the current user error cause, and displaying the error cause analysis result on the answer page of the target question, it also includes: Displaying the second error cause knowledge point and the third error cause knowledge point in the bottom area of the question answering area; In response to a user clicking on the second error cause knowledge point or the third error cause knowledge point in the bottom area, the knowledge analysis content corresponding to the error cause knowledge point clicked by the user is displayed on the question answering page, wherein the knowledge analysis content includes one or more of video content or graphic content.
8. The method according to claim 3, characterized in that After determining the error cause analysis result of the draft paper file according to the current user error cause, the method further includes: The current user error cause generated by the error cause analysis model, the error cause analysis result and the first error cause knowledge point are updated into the error cause pool.
9. The method according to claim 1, characterized in that: Before responding to the user's uploading operation of at least one draft paper file for at least one target topic, the method further includes: Obtain one or more of the standard answer to the target question or the user's answer to the target question, and update them as part of the draft paper file.
10. The method according to claim 1, characterized in that Before obtaining at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps, the method further includes: Acquire historical question-solving information, wherein the historical question-solving information includes historical user error causes, historical error cause classifications, and historical error cause analysis results corresponding to the answer contents of different users for different target questions of different subjects; The historical user error causes, historical error cause classifications, and historical error cause analysis results are input into a large language model, and the large language model is trained in a supervised manner until the output results of the large language model meet the preset results of the historical user error causes, historical error cause classifications, and historical error cause analysis results, and the trained large language model is used as the error cause analysis model.
11. The method according to claim 4, characterized in that Also includes: Respectively determine at least one candidate error cause knowledge point corresponding to the current user error cause, the first error cause knowledge point, the second error cause knowledge point, and the associated topics related to the third error cause knowledge point; The related topics are pushed.
12. An artificial intelligence-based error cause identification device, characterized in that: include: An operation response module, used to respond to a user's upload operation of at least one draft paper file for at least one target question, wherein the draft paper file includes one or more of the solution ideas or solution steps generated by the user for the target question; An error cause acquisition module, used to acquire at least one current user error cause generated by the trained error cause analysis model based on at least one of the problem-solving ideas or problem-solving steps; The result generation module is used to determine the error cause analysis result of the draft paper file according to the current user error cause, and display the error cause analysis result on the answer page of the target question.
13. The device according to claim 12, characterized in that The result generation module comprises: The error cause pool positioning unit is used to locate the error cause pool where the target topic is located, wherein the error cause pool is a directed graph with target topic, error cause classification, candidate user error cause, candidate error cause analysis and candidate error cause knowledge points as different hierarchical nodes and subordinate relationships between nodes at each level as directed edges, and the error cause pool includes different candidate user error causes under the same target topic and error cause analysis of each candidate user error cause; an error cause classification determination unit, configured to determine an error cause classification of the current user error cause if the error cause of the current user for the target question exists in the error cause pool; wherein the error cause classification includes knowledge point error causes and non-knowledge point error causes; A first result determination unit is configured to, when the error cause of the current user error cause is classified as the knowledge point error cause, identify at least one candidate error cause knowledge point corresponding to the current user error cause in the error cause pool, and use the identified candidate error cause knowledge point as the error cause analysis result of the draft paper file; The second result determination unit is used to identify, when the error cause of the current user error is classified as the non-knowledge point error cause, a candidate error cause analysis corresponding to the current user error cause in the error cause pool as the error cause analysis result of the draft paper file.
14. The device according to claim 12, characterized in that The result generation module further includes: an error cause classification acquisition unit, configured to acquire an error cause classification of the current user error cause generated by the error cause analysis model if the current user error cause of the target question does not exist in the error cause pool; A third result determination unit is used to obtain an error cause analysis result of the draft paper file generated by the error cause analysis model based on the current user error cause when the error cause is classified as a non-knowledge point error cause; a knowledge point determination unit, configured to obtain, when the error cause is classified as a knowledge point error cause, a first error cause knowledge point generated by the error cause analysis model and having knowledge relevance to the current user error cause, wherein the knowledge relevance is determined based on the current user error cause and the subordinate relationship between the target question and the question knowledge point; The fourth result determination unit is used to use the candidate error cause knowledge point that matches the first error cause knowledge point in the error cause knowledge point map as the error cause analysis result of the draft paper file.
15. The device according to claim 14, characterized in that The knowledge point matching unit is specifically used for: Acquire a candidate error cause knowledge point that matches the first error cause knowledge point from the error cause knowledge point map as a second error cause knowledge point; Acquire historical answer data of the target question, wherein the historical answer data includes a first set of knowledge points learned by learners who answered the target question correctly and a second set of knowledge points learned by learners who answered the target question incorrectly; At least one candidate error cause knowledge point that exists in the first knowledge point set but does not exist in the second knowledge point set is used as a third error cause knowledge point; The second error cause knowledge point and the third error cause knowledge point are used as the error cause analysis result of the draft paper file.
16. The device according to claim 13, characterized in that The non-knowledge point error causes include at least one of unclear examination of the question, carelessness and transcription errors.
17. The device according to claim 12, characterized in that The operation response module is specifically used for: In response to the user's answering operation on the target question, when the electronic draft paper is detected, the electronic draft paper is used as the draft paper file; In response to a user's upload operation on the electronic draft paper.
18. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the artificial intelligence-based error cause identification method described in any one of claims 1-11.
19. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the artificial intelligence-based error cause identification method described in any one of claims 1-11 when executed.
20. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the artificial intelligence-based error cause identification method according to any one of claims 1 to 11.
