An ai intelligent question analysis and interaction method and system

Abstract

The application discloses an AI intelligent question analysis and interaction method and system and relates to the field of question analysis. A question sentence, an answer sentence and an answer sentence are acquired. Based on the question sentence, the answer sentence and the answer sentence, an error step and a missing step are segmented, and a keyword sequence is obtained. Through a question recommendation network, based on the keyword sequence, the relationship between the question and the error step and the missing step is detected, and a student interaction question is obtained. According to the correlation relationship between the keywords of each answer step and the keywords of the question, the importance of each answer step for solving the question can be known, and a step importance value is obtained. According to the step importance value, the question recommendation network is constructed, and the technical effect of more accurately recommending the question with easy-to-miss points to the student is achieved.

Description

Technical Field

[0001] This invention relates to the field of question analysis, and more specifically, to an AI-powered intelligent question analysis and interaction method and system. Background Technology

[0002] Currently, with the rapid development of artificial intelligence technology, especially the maturity of natural language processing (NLP) and machine learning algorithms, AI is being applied more and more widely in the field of education. There are already various online education platforms and intelligent question bank systems on the market, whose core functions typically include: OCR-based text recognition, simple answer matching (such as automatic grading of objective questions), and recommendations based on students' historical answer accuracy.

[0003] However, significant shortcomings and deficiencies remain when dealing with subjective questions (such as essay questions and mathematical proof questions) and complex interactive scenarios. Most existing technologies are limited to binary judgments of "right or wrong answers," lacking in-depth semantic analysis of the questions themselves. For example, they cannot effectively identify and accurately pinpoint students' knowledge gaps or logical breaks in their problem-solving process.

[0004] Some studies use deep learning for semantic mining, but they do not mine content in different directions based on the different errors in students' answers, nor do they mine the correlation between questions and answers.

[0005] Therefore, how to provide a method that can perform in-depth intelligent decomposition of questions and achieve adaptive, generative, and guided interaction on this basis has become a pressing technical challenge. Summary of the Invention

[0006] The purpose of this invention is to provide an AI-powered intelligent question analysis and interaction method and system to solve the aforementioned problems existing in the prior art.

[0007] In a first aspect, embodiments of the present invention provide an AI-powered intelligent question analysis and interaction method, comprising:

[0008] The system retrieves the question statement, the answer statement, and the correct answer statement; the question statement represents the statement in which the student answers a question; the answer statement represents the statement in which the student responds; and the correct answer statement represents the statement in which the standard answer is given.

[0009] Based on the question statement, answer statement, and solution statement, incorrect steps and missing steps are segmented to obtain a keyword sequence; the keyword sequence includes an incorrect keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0010] Based on the question statement, answer statement, and response statement, incorrect steps and missing steps are segmented to obtain a keyword sequence; the keyword sequence includes an incorrect keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0011] Optionally, the topic recommendation network includes a feature extraction network, a temporal convolutional network, and a causal convolutional network.

[0012] Optionally, the step of segmenting based on the question statement, answer statement, and response statement, detecting keywords, and obtaining a keyword sequence includes:

[0013] The keywords in the question statement, the answer statement, and the answer statement are detected respectively, and multiple question keywords, multiple answer keywords, and multiple answer keywords are obtained respectively.

[0014] Based on the answer statement and the response statement, step judgment is performed to obtain m steps to be judged; the steps to be judged are missing steps or incorrect steps;

[0015] Based on m steps to be judged, m importance values ​​are obtained; the importance value of a step represents the degree of correlation between the knowledge point in a step and the knowledge point of the question.

[0016] Based on m importance values, m steps to be judged, and multiple question keywords, a keyword sequence is obtained; the keyword sequence contains n incorrect keyword sequences and p missing keyword sequences.

[0017] Optionally, the keyword sequence obtained based on m importance values, m steps to be judged, and multiple question keywords includes:

[0018] Sort the answer keywords in the incorrect steps according to their order in the incorrect steps to obtain the incorrect keyword sequence;

[0019] Based on the missing steps and multiple question keywords, a sequence of missing keywords is obtained;

[0020] m steps to be judged correspond to n erroneous keyword sequences and p missing keyword sequences;

[0021] Where n + p = m;

[0022] Sort the n erroneous keyword sequences and p missing keyword sequences according to their importance values ​​from smallest to largest to obtain the keyword sequence.

[0023] Optionally, the step of recommending questions based on the keyword sequence through a question recommendation network to obtain student interaction questions includes:

[0024] The keywords in the keyword sequence are input into the feature extraction network to extract features and obtain the first feature vector;

[0025] The first feature vector is input into the temporal convolutional network according to the index order in the missing keyword sequence or the erroneous keyword sequence to obtain m second feature vectors;

[0026] According to the importance value of each step from smallest to largest, the m second feature vectors are input into the causal convolutional network to obtain the feature vector of the keyword to be learned.

[0027] Obtain multiple keyword feature vectors for the questions; the keyword feature vectors represent the feature vectors of keywords in the questions stored in the database;

[0028] The feature vector of the keyword to be learned is matched with the feature vectors of multiple question keywords to obtain interactive questions for students.

[0029] Optionally, the step of obtaining a missing keyword sequence based on the missing step and multiple question keywords includes:

[0030] Obtain multiple storage keywords and their corresponding sets of storage knowledge points; the set of storage knowledge points includes a set of multiple knowledge points associated with the storage keywords; the storage keywords represent keywords that have already been stored.

[0031] In the missing keywords step, answer keywords are extracted as missing keywords; the missing keywords refer to the keywords missing from the answer statement.

[0032] Based on the missing keywords, obtain the set of missing knowledge points; the set of missing knowledge points represents the set of stored knowledge points associated with the missing keywords;

[0033] Set the knowledge points in the set of missing knowledge points as missing knowledge points;

[0034] The keywords corresponding to the set of stored knowledge points containing the missing knowledge points are used as the second missing keywords; the number of the second missing keywords is greater than or equal to the number of missing keywords.

[0035] Multiple title keywords are matched with multiple second missing keywords to obtain multiple third missing keywords; the third missing keywords represent the matched title keywords.

[0036] Arrange the multiple missing keywords according to their order in the question statement to obtain the missing keyword sequence.

[0037] Optionally, based on the answer statement and the response statement, the step judgment is performed to obtain m steps to be judged, including:

[0038] Based on the answer statement and the response statement, a matching vector is obtained; the matching vector represents the relationship between the steps in the answer statement and the response statement.

[0039] Take the index of the matching vector with a value of 0 as the second index; take the index of the second index with a non-zero value before it as the first index, and the index of the second index with a non-zero value after it as the third index;

[0040] If the difference between the values ​​of the first subscript and the third subscript is greater than 1, the answer step corresponding to the second subscript is set as an incorrect step;

[0041] If the difference between the values ​​of the first and third subscripts is equal to 1, set the answer step corresponding to the second subscript as a missing step.

[0042] Optionally, obtaining the matching vector based on the answer statement and the response statement includes:

[0043] Based on the answer statement, obtain the answer step vector; the first index of the answer step vector is 0, and the values ​​corresponding to the other indices are multiple answer steps arranged in sequence;

[0044] The answer statement is divided into multiple answer steps to obtain an answer step vector; the first index of the answer step vector is 0, and the values ​​of the other indices are multiple answer steps arranged in sequence.

[0045] The matching algorithm is used to determine whether the multiple answer steps match the multiple response steps.

[0046] Associate the index of the answer step vector with the index of the corresponding matching vector; one index of the matching vector corresponds to one answer step;

[0047] If a match is found, the value in the matching vector is set to the index of the matching answer step in the answer step vector; otherwise, the value in the matching vector is set to 0, thus obtaining the matching vector.

[0048] Optionally, the step of obtaining m importance values ​​based on m steps to be judged includes:

[0049] The intersection of the answer keywords and question keywords corresponding to the missing or incorrect steps is used to obtain a similarity set.

[0050] Based on the similarity set, the importance value of the step is obtained;

[0051] For each of the m steps to be judged, m importance values ​​are obtained for each step.

[0052] The method for calculating the importance value of the aforementioned steps is as follows:

[0053]

[0054] Where y represents the importance value of the step, and x represents the number of elements in the similar set.

[0055] Secondly, embodiments of the present invention provide an AI-powered intelligent question analysis and interaction system, comprising:

[0056] The acquisition module is used to acquire question statements, answer statements, and standard answer statements; the question statements represent the statements in which students answer questions; the answer statements represent the statements in which students answer questions; and the standard answer statements represent the statements in which standard answers are provided.

[0057] The keyword extraction module is used to segment erroneous steps and missing steps based on the question statement, answer statement, and response statement to obtain a keyword sequence; the keyword sequence includes an erroneous keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0058] The question recommendation module is used to segment erroneous steps and missing steps based on the question statement, answer statement, and answer statement to obtain a keyword sequence; the keyword sequence includes an erroneous keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0059] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0060] This invention also provides an AI-powered intelligent question analysis and interaction method and system.

[0061] In this application, the answer statement and the response statement are each divided into multiple steps. For declarative questions with fixed answer structures, a process is performed. Based on the answer statement and response statement, it is determined whether there are erroneous or missing steps. If there is an erroneous step, the keywords indicating the error can be directly obtained. If there is a missing step, to assess the student's understanding of the knowledge points and thus determine the reason for the missing step, the knowledge points of the missing answer step are associated with the knowledge points of the question, and the associated keywords in the question are used as missing keywords. Based on the association between the keywords of each answer step and the keywords of the question, the importance of each answer step in solving the problem and the importance of the knowledge points the student needs to learn can be determined, thus obtaining a step importance value. Based on the step importance value and the order of keywords in the question and response statements, a question recommendation network is constructed, achieving a more accurate technical effect of recommending questions that students are prone to making mistakes and losing points on. Attached Figure Description

[0062] Figure 1 This is a flowchart of an AI-powered intelligent question analysis and interaction method provided in an embodiment of the present invention. Detailed Implementation

[0063] The present invention will now be described in detail with reference to the accompanying drawings.

[0064] This application relates to an AI-powered intelligent question analysis and interaction method and system for questions where the order of the steps to answer the question is the same as the answer itself.

[0065] Example 1:

[0066] like Figure 1 As shown, this embodiment of the invention provides an AI-powered intelligent question analysis and interaction method, the method comprising:

[0067] S101: Obtain the question statement, the answer statement, and the correct answer statement; the question statement represents the statement in which the student answers the question; the answer statement represents the statement in which the student answers the question; the correct answer statement represents the statement in which the standard answer is given.

[0068] S102: Based on the question statement, answer statement, and answer statement, segment the incorrect steps and missing steps to obtain a keyword sequence; the keyword sequence includes an incorrect keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0069] The keywords mentioned here refer to keywords related to knowledge point detection. For example, if the knowledge point is "geometry", the keywords include "line segment", "triangle", "curve", and "angle".

[0070] The keyword sequence represents the keywords corresponding to knowledge points that students have not mastered.

[0071] S104: Based on the question statement, answer statement, and answer statement, segment the incorrect steps and missing steps to obtain a keyword sequence; the keyword sequence includes an incorrect keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0072] In this embodiment, the question recommendation network uses a causal convolutional network to extract feature information of knowledge points that students are prone to making mistakes. By comparing and matching, it determines the similarity with questions stored in the database to find interactive questions for students.

[0073] Optionally, the keyword sequence obtained based on m importance values, m steps to be judged, and multiple question keywords includes:

[0074] The keywords in the question statement, answer statement, and answer statement are detected respectively, and multiple question keywords, multiple answer keywords, and multiple answer keywords are obtained respectively.

[0075] In this embodiment, the KeyBERT algorithm is used for keyword extraction.

[0076] Based on the answer statement and the response statement, step judgment is performed to obtain m steps to be judged; the steps to be judged are missing steps or incorrect steps.

[0077] Where m is a positive integer.

[0078] In this context, "missing steps" refers to steps that are missing from the answer statement compared to the original answer statement; "incorrect steps" refers to steps that are incorrectly answered in the answer statement compared to the original answer statement. The missing steps can be used to determine if a student has not written them and to provide a reminder. "Incorrect steps" indicate that the student has completely failed to grasp the corresponding knowledge point.

[0079] The above method addresses the issue that students may omit or make mistakes when answering declarative or problem-solving questions, resulting in lost marks. Therefore, instead of judging the entire question, the evaluation is performed by analyzing individual steps.

[0080] Based on m steps to be judged, m importance values ​​are obtained; the importance value of a step represents the degree of correlation between the knowledge point in a step and the knowledge point of the question.

[0081] Using the above method, if the missing or incorrect steps contain knowledge points that the question is testing, then the steps are more important.

[0082] Based on m importance values, m steps to be judged, and multiple question keywords, a keyword sequence is obtained; the keyword sequence contains n incorrect keyword sequences and p missing keyword sequences.

[0083] Wherein, the error keyword sequence represents the sequence in the question statement associated with the error step; the missing keyword sequence represents the sequence in the question statement associated with the missing step.

[0084] Optionally, the keyword sequence obtained based on m importance values, m steps to be judged, and multiple question keywords includes:

[0085] Sort the answer keywords in the incorrect steps according to their order in the incorrect steps to obtain the incorrect keyword sequence.

[0086] Using the method described above, since each step indicates that the student has not fully grasped the corresponding knowledge point, the keywords in that step are recorded as the knowledge points that the student needs to master.

[0087] Based on the missing steps and multiple title keywords, a sequence of missing keywords is obtained.

[0088] For each of the m steps to be judged, n sequences of incorrect keywords and p sequences of missing keywords are obtained.

[0089] Where n + p = m.

[0090] Where n, p, and m are positive integers.

[0091] Sort the n erroneous keyword sequences and p missing keyword sequences according to their importance values ​​from smallest to largest to obtain the keyword sequence.

[0092] Optionally, the topic recommendation network includes a feature extraction network, a temporal convolutional network, and a causal convolutional network.

[0093] Optionally, the step of recommending questions based on the keyword sequence through a question recommendation network to obtain student interaction questions includes:

[0094] The keywords in the keyword sequence are input into the feature extraction network to extract features and obtain the first feature vector.

[0095] The first feature vector represents the feature information of the keywords of the knowledge point to be learned.

[0096] In this embodiment, the feature extraction network is a fully connected neural network (FCNN). The input to the feature extraction network is the numerical form of the keywords converted from their textual form.

[0097] The first feature vector is input into a temporal convolutional network (TCN) according to the index order in the missing keyword sequence or the erroneous keyword sequence to obtain m second feature vectors.

[0098] The second feature vector represents the feature information of multiple keywords related to a step.

[0099] The reason for using a Temporal Convolutional Network (TCN) in this way is that the keywords have a sequential arrangement in a step.

[0100] Following the principle of importance from smallest to largest, the m second feature vectors are input into the Causal Convolutional Network to obtain the feature vectors of the keywords to be learned.

[0101] The feature vector of the keyword to be learned represents the feature information of the keyword corresponding to the knowledge points that students have not yet mastered in a question.

[0102] Using the method described above, due to the characteristics of causal convolutional networks, the data input at the end actually retains more feature information than the data input at the beginning. Therefore, the input is arranged in ascending order of importance. Since the keyword sequence has already been sorted during construction, it can be directly input into the causal convolutional network according to its index.

[0103] Obtain multiple keyword feature vectors for the questions; the keyword feature vectors represent the feature vectors of keywords in the questions stored in the database.

[0104] The keyword feature vector in the title is calculated using the same method as the keyword feature vector, which involves using a feature extraction network, a causal convolutional network, and a temporal convolutional network for judgment.

[0105] The feature vector of the keyword to be learned is matched with the feature vectors of multiple question keywords to obtain interactive questions for students.

[0106] In this embodiment, similarity is calculated using Euclidean distance for matching. If the similarity is greater than the similarity threshold, the feature vector of the keyword to be learned is set to match the feature vector of the keyword in the question, and the question corresponding to the feature vector of the keyword in the question is taken as the interactive question for students. In this embodiment, the similarity threshold is 0.85.

[0107] Optionally, the step of obtaining the missing keyword sequence based on the missing steps and the question statement includes:

[0108] Obtain multiple storage keywords and their corresponding sets of storage knowledge points; the set of storage knowledge points includes a set of multiple knowledge points associated with the storage keywords; the storage keywords represent the keywords that have already been stored.

[0109] Based on the missing steps, extract the answer keywords to obtain the missing keywords; the missing keywords refer to the keywords missing from the answer statement;

[0110] Based on the missing keywords, obtain the set of missing knowledge points; the set of missing knowledge points represents a collection of multiple knowledge points associated with the keywords.

[0111] In this embodiment, multiple keywords are stored in the database, and each keyword is associated with multiple knowledge points. Therefore, obtaining a missing keyword allows for the direct extraction of the corresponding associated knowledge points.

[0112] Set the knowledge points in the set of missing knowledge points as missing knowledge points;

[0113] The keywords corresponding to the set of stored knowledge points containing the missing knowledge points are used as the second missing keywords; the number of the second missing keywords is greater than or equal to the number of missing keywords.

[0114] In this case, because the missing keywords correspond to multiple missing knowledge points, and since the database contains more keywords containing missing knowledge points than the missing keywords themselves, all keywords related to the missing knowledge points are extracted. These keywords are then matched with multiple question keywords. Compared to directly matching question keywords with missing keywords, this method retrieves keywords that differ but contain the same knowledge points.

[0115] Multiple title keywords are matched with multiple second missing keywords to obtain multiple third missing keywords; the third missing keywords represent the matched title keywords.

[0116] The number of the third missing keywords is less than or equal to the number of the second missing keywords.

[0117] Arrange the multiple missing keywords according to their order in the question statement to obtain the missing keyword sequence.

[0118] Optionally, the step determination based on the answer statement and the response statement to obtain missing steps and incorrect steps includes:

[0119] Based on the answer statement and the response statement, a matching vector is obtained; the matching vector represents the relationship between the steps in the answer statement and the response statement.

[0120] Take the index of the matching vector with a value of 0 as the second index; take the index of the second index with a non-zero value before it as the first index, and the index of the second index with a non-zero value after it as the third index;

[0121] If the difference between the values ​​of the first and third subscripts is greater than 1, the answer step corresponding to the second subscript is set as an incorrect step.

[0122] If the answer step vector is [0,A,B,C] and the response step vector is [0,A,D,C], then the matching vector is [0,1,0,3]. Since 3-1=2, which is greater than 1, it indicates that step D is an incorrect step. If the answer step vector is [0,A,B,C] and the response step vector is [0,A,C], then the matching vector is [0,1,0,2]. Since 2-1=1, it indicates that step B is missing.

[0123] If the difference between the values ​​of the first and third subscripts is equal to 1, set the answer step corresponding to the second subscript as a missing step.

[0124] As shown in Table 1, if the second answer step differs from the second response step, the second response step is set as a misaligned step.

[0125] Optionally, obtaining the matching vector based on the answer statement and the response statement includes:

[0126] Based on the answer statement, obtain the answer step vector; the first index of the answer step vector is 0, and the values ​​corresponding to the other indices are multiple answer steps arranged in sequence.

[0127] The purpose of constructing the first special answer step vector with index 0 is to build the values ​​in the matching vector. Since indices start from 0, if a match is found, the value in the matching vector is set to the corresponding index; otherwise, it's set to 0. To prevent the step with index 0 from matching but being filled into the matching vector with a value of 0 (representing a mismatch), a value of 0 that doesn't match any answer step is filled at index 0 when constructing the answer step vector. The index corresponding to the first answer step is 1.

[0128] In this embodiment, the answer step vector is [0, A, B, C], where A represents the content of the first answer step, B represents the content of the second answer step, and C represents the content of the third answer step.

[0129] Wherein, the answer step refers to the step after splitting the answer statement; the answer step is the step of marking.

[0130] The answer statement is divided into multiple answer steps to obtain an answer step vector; the first index of the answer step vector is 0, and the values ​​corresponding to the other indices are multiple answer steps arranged in sequence.

[0131] In this embodiment, the answer step vector is [0, A, D, C]. A represents the content of the first answer step, i.e., the content of the first step of the answer. D represents the content of the second answer step, i.e., the content of the second step of the answer, and C represents the content of the third answer step, i.e., the content of the third step of the answer.

[0132] In this embodiment, which is a math problem, the answer statement and the response statement are separated by an equal sign.

[0133] In another embodiment, for a Chinese language question, the answer statement and the reply statement are separated by a period.

[0134] A matching algorithm is used to determine whether the multiple answer steps match the multiple response steps.

[0135] In this embodiment, the search is performed sequentially, with each step of calculating one answer being matched against multiple answer steps. The matching algorithm is a string matching algorithm, and the entire mathematical problem is directly matched as a whole.

[0136] In another embodiment, it is used to determine whether the characters of the keywords in the Chinese language statement title are the same.

[0137] If a match is found, the value in the matching vector is set to the index of the matching answer step in the answer step vector; otherwise, the value in the matching vector is set to 0, thus obtaining the matching vector.

[0138] In this embodiment, if the answer step vector is [0,A,B,C] and the response step vector is [0,A,D,C], then the matching vector is [0,1,0,3]. If the value is 0, it indicates that the i-th answer step at index i is incorrect. If the value is not 0, the value of j at index i indicates that the i-th answer step matches the j-th response step. i and j are greater than 0. If the answer step vector is [0,A,B,C] and the response step vector is [0,A,C], then the matching vector is [0,1,0,2]. If the value is 0, it indicates that the i-th answer step at index i is missing. If the value is not 0, the value of j at index i indicates that the i-th answer step matches the j-th response step.

[0139] Optionally, for the m steps to be judged, m importance values ​​are obtained, including:

[0140] The intersection of the answer keywords and question keywords corresponding to the missing or erroneous steps is used to obtain a similarity set.

[0141] Wherein, the number of elements in the similar set is greater than or equal to 0.

[0142] Based on the similarity set, the importance value of the step is obtained.

[0143] For each of the m steps to be judged, m importance values ​​are obtained.

[0144] The method for calculating the importance value of the aforementioned steps is as follows:

[0145]

[0146] Where y represents the importance value of the step, and x represents the number of elements in the similar set.

[0147] Using the above method, since the number of elements in the similar set is greater than or equal to 0, the logarithm is calculated by adding one to the number of elements. This ensures that the number of elements whose keywords corresponding to the missing or erroneous steps are the same as the keywords corresponding to the question is 0. In this case, the importance value of the step is 0. The more elements there are, the greater the importance value of the step, and the more elements there are, the slower the change.

[0148] Example 2:

[0149] Based on the above-described AI-powered intelligent question analysis and interaction methods, this invention also provides an AI-powered intelligent question analysis and interaction system, the system comprising:

[0150] The acquisition module is used to acquire question statements, answer statements, and standard answer statements; the question statements represent the statements in which students answer questions; the answer statements represent the statements in which students answer questions; and the standard answer statements represent the statements in which standard answers are provided.

[0151] The keyword extraction module is used to segment erroneous steps and missing steps based on the question statement, answer statement, and response statement to obtain a keyword sequence; the keyword sequence includes an erroneous keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

[0152] The question recommendation module is used to recommend questions based on the keyword sequence through a question recommendation network, thereby obtaining student interactive questions; the student interactive questions represent knowledge points that students are prone to making mistakes on.

[0153] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0154] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0155] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. An AI-powered intelligent question analysis and interaction method, characterized in that, include: Retrieve the question statement, the answer statement, and the response statement; Based on the question statement, answer statement, and solution statement, incorrect steps and missing steps are segmented to obtain a keyword sequence; the keyword sequence includes an incorrect keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points. By using a question recommendation network and based on the keyword sequence, the relationship between questions and incorrect or missing steps is detected to obtain interactive questions for students.

2. The AI-powered intelligent question analysis and interaction method according to claim 1, characterized in that, The topic recommendation network includes a feature extraction network, a temporal convolutional network, and a causal convolutional network.

3. The AI-powered intelligent question analysis and interaction method according to claim 1, characterized in that, The segmentation step based on the question statement, answer statement, and response statement, detecting keywords, and obtaining a keyword sequence includes: The keywords in the question statement, the answer statement, and the answer statement are detected respectively, and multiple question keywords, multiple answer keywords, and multiple answer keywords are obtained respectively. Based on the answer statement and the response statement, step judgment is performed to obtain m steps to be judged; the steps to be judged are missing steps or incorrect steps; Based on m steps to be judged, m importance values ​​are obtained; the importance value of a step represents the degree of correlation between the knowledge point in a step and the knowledge point of the question. Based on m importance values, m steps to be judged, and multiple question keywords, a keyword sequence is obtained; the keyword sequence contains n incorrect keyword sequences and p missing keyword sequences.

4. The AI-powered intelligent question analysis and interaction method according to claim 3, characterized in that, Based on m importance values, m steps to be judged, and multiple question keywords, a keyword sequence is obtained, including: Sort the answer keywords in the incorrect steps according to their order in the incorrect steps to obtain the incorrect keyword sequence; Based on the missing steps and multiple question keywords, a sequence of missing keywords is obtained; m steps to be judged correspond to n erroneous keyword sequences and p missing keyword sequences; Where n + p = m; Sort the n erroneous keyword sequences and p missing keyword sequences according to their importance values ​​from smallest to largest to obtain the keyword sequence.

5. The AI-powered intelligent question analysis and interaction method according to claim 2, characterized in that, The step of recommending questions based on the keyword sequence through a question recommendation network to obtain student interaction questions includes: The keywords in the keyword sequence are input into the feature extraction network to extract features and obtain the first feature vector; The first feature vector is input into the temporal convolutional network according to the index order in the missing keyword sequence or the erroneous keyword sequence to obtain m second feature vectors; According to the importance value of each step from smallest to largest, the m second feature vectors are input into the causal convolutional network to obtain the feature vector of the keyword to be learned. Obtain multiple keyword feature vectors for the questions; the keyword feature vectors represent the feature vectors of keywords in the questions stored in the database; The feature vector of the keyword to be learned is matched with the feature vectors of multiple question keywords to obtain interactive questions for students.

6. The AI-powered intelligent question analysis and interaction method according to claim 4, characterized in that, The missing keyword sequence is obtained based on the missing steps and multiple question keywords, including: Obtain multiple storage keywords and their corresponding sets of storage knowledge points; the set of storage knowledge points includes a set of multiple knowledge points associated with the storage keywords; the storage keywords represent keywords that have already been stored. In the missing keywords step, answer keywords are extracted as missing keywords; the missing keywords refer to the keywords missing from the answer statement. Based on the missing keywords, obtain the set of missing knowledge points; the set of missing knowledge points represents the set of stored knowledge points associated with the missing keywords; Set the knowledge points in the set of missing knowledge points as missing knowledge points; The keywords corresponding to the set of stored knowledge points containing the missing knowledge points are used as the second missing keywords; the number of the second missing keywords is greater than or equal to the number of missing keywords. Multiple title keywords are matched with multiple second missing keywords to obtain multiple third missing keywords; the third missing keywords represent the matched title keywords. Arrange the multiple missing keywords according to their order in the question statement to obtain the missing keyword sequence.

7. The AI-powered intelligent question analysis and interaction method according to claim 3, characterized in that, The step judgment is performed based on the answer statement and the response statement to obtain m steps to be judged, including: Based on the answer statement and the response statement, a matching vector is obtained; the matching vector represents the relationship between the steps in the answer statement and the response statement. Take the index of the matching vector with a value of 0 as the second index; take the index of the second index with a non-zero value before it as the first index, and the index of the second index with a non-zero value after it as the third index; If the difference between the values ​​of the first subscript and the third subscript is greater than 1, the answer step corresponding to the second subscript is set as an incorrect step; If the difference between the values ​​of the first and third subscripts is equal to 1, set the answer step corresponding to the second subscript as a missing step.

8. The AI-powered intelligent question analysis and interaction method according to claim 7, characterized in that, The process of obtaining a matching vector based on the answer statement and the response statement includes: Based on the answer statement, obtain the answer step vector; the first index of the answer step vector is 0, and the values ​​corresponding to the other indices are multiple answer steps arranged in sequence; The answer statement is divided into multiple answer steps to obtain an answer step vector; the first index of the answer step vector is 0, and the values ​​of the other indices are multiple answer steps arranged in sequence. The matching algorithm is used to determine whether the multiple answer steps match the multiple response steps. Associate the index of the answer step vector with the index of the corresponding matching vector; one index of the matching vector corresponds to one answer step; If a match is found, the value in the matching vector is set to the index of the matching answer step in the answer step vector; otherwise, the value in the matching vector is set to 0, thus obtaining the matching vector.

9. The AI-powered intelligent question analysis and interaction method according to claim 3, characterized in that, The m steps to be judged are used to obtain m importance values, including: The intersection of the answer keywords and question keywords corresponding to the missing or incorrect steps is used to obtain a similarity set. Based on the similarity set, the importance value of the step is obtained; For each of the m steps to be judged, m importance values ​​are obtained for each step. The method for calculating the importance value of the aforementioned steps is as follows: Where y represents the importance value of the step, and x represents the number of elements in the similar set.

10. An AI-powered intelligent question analysis and interaction system, characterized in that, include: The acquisition module is used to acquire the question statement, the answer statement, and the response statement; the question statement represents the statement used by the student to answer the question. The answer statement refers to the statement made by the student in response; the answer statement refers to the statement of the standard answer. The keyword extraction module is used to segment erroneous steps and missing steps based on the question statement, answer statement, and response statement to obtain a keyword sequence; the keyword sequence includes erroneous keyword sequences and missing keyword sequences. The keyword sequence contains multiple keywords arranged according to knowledge points; The question recommendation module is used to segment erroneous steps and missing steps based on the question statement, answer statement, and answer statement to obtain a keyword sequence; the keyword sequence includes an erroneous keyword sequence and a missing keyword sequence; the keyword sequence contains multiple keywords arranged according to knowledge points.

Images