Personalized test question recommendation method and system based on question bank

Abstract

The invention discloses a personalized test question recommendation method and system based on a question bank, and relates to the technical field of test question recommendation. A personalized test question recommendation system based on a question bank comprises a user test question recommendation ability analysis module, a test question data analysis module, a user test question recommendation emotion analysis module and a user test question recommendation prediction module. According to the method, the answer data and the behavior data of the target user are acquired, and the ability analysis and emotion analysis models are combined, so that accurate personalized recommendation for the user can be realized, and the most suitable test questions can be recommended according to the ability increase and emotion fluctuation of the user, so that the learning efficiency is improved; by converting the test question data into the analysis test question vector and analyzing the analysis test question vector by using the personalized test question analysis model, the mastering condition of the user on each knowledge point can be comprehensively evaluated, not only is the answering accuracy of the user concerned, but also the specific performance of the user in the answering process is considered, so that more comprehensive ability evaluation is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of test question recommendation, and in particular to a method and system for recommending personalized test questions based on a question bank. Background Art

[0002] Test question recommendations usually recommend test questions that are suitable for students' personal learning needs and ability levels by analyzing their learning behaviors, historical performance, and knowledge points. This process requires accurate educational measurement of students' cognitive levels. Traditional educational measurement methods mainly reflect students' mastery of knowledge points through test scores, but this method has limitations. For example, even students with the same test scores may have different levels of mastery of knowledge points. Test scores alone cannot accurately reflect students' true ability levels, nor can they distinguish differences in knowledge point mastery among different students. In addition, if students do not have a solid grasp of a certain knowledge point, it may affect their understanding and construction of the entire knowledge system. Whether students can make up for their learning shortcomings and improve their grades depends largely on whether the test questions in subsequent practice can effectively help them consolidate and improve. In this process, personalized test question recommendations play a vital role. Summary of the invention

[0003] The present invention aims to provide a personalized test question recommendation method and system based on a question bank, so as to recommend test questions that are more suitable for the current learning stage to users.

[0004] A personalized test question recommendation method based on a question bank comprises the following steps: Step S1: Obtain the target user's stage answer data; the target user's stage answer data contains N pieces of target user's answered test data T n and the target user answer option data P n , n=1, 2, ..., N; N is the total number of questions answered by the target user in the current stage; based on the data of questions answered by the target user T n , Target user answer option data P n And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; Get the target user's answered test data T n Specific steps: For any test question, obtain the test question data and convert the test question data into an analysis test question vector X, X=[X1, X2, …, X i , …, X I ], X irepresents the specific information of the i-th question feature in the analysis question vector X, where I is the total number of question features in the question data; the analysis question vector is input into the personalized question analysis model for analysis, and the question analysis ability vector Y is obtained, where Y=[Y1, Y2, …, Y i , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; Step S2: Obtain the target user's stage answering behavior data, which includes the target user's stay time S for answering questions. n Interaction data H with target users n ; Based on the target user's stage answering behavior data and the user recommended test question sentiment analysis model, the target user's sentiment fluctuation factor E is obtained n The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; Step S3: According to the target user's ability growth factor Z n , Target user emotion fluctuation factor E n The user test question recommendation prediction model is analyzed to obtain a target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, corresponding test questions are recommended for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which are used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities.

[0005] As a preferred technical solution of the present invention, the personalized test question analysis model in step S1 includes a feature recognition layer, a feature analysis layer C i And feature result output layer; The feature recognition layer is used to analyze all X in the question vector X. i Perform feature recognition and obtain feature label B i ; According to feature label B i X i Send to the corresponding feature analysis layer C i ; Feature Analysis Layer C iFor X i Perform feature analysis and obtain Y i ; The feature result output layer is used to convert all Y i Combine them to get the question analysis ability vector Y, Y=[Y1, Y2,…, Y i , …, Y I ]; Training feature analysis layer C i The specific steps include: According to the setting feature label B i Collecting several groups of test question feature training samples corresponding to test question features; each group of test question feature training samples contains a group of specific information corresponding to test question features and test question ability analysis data; combining several groups of test question feature training samples to obtain a test question feature training set; According to the test feature training set, the feature analysis layer C i Model training is performed with the test ability analysis data as the target to obtain the initial feature analysis layer C i '; Initial feature analysis layer C i 'Evaluate the model and get the initial feature analysis layer C i 'Model evaluation results; if the initial feature analysis layer C i 'If the model evaluation result is passed, the initial feature analysis layer C i 'As feature analysis layer C i , deployed to the personalized test question analysis model, otherwise, continue model training; Traverse and train all feature analysis layers C i , until the model training is completed.

[0006] As a preferred technical solution of the present invention, the user recommendation test question ability analysis model in step S1 includes a score judgment layer, an ability increment calculation layer and an ability factor output layer; The scoring judgment layer is used to judge the target user's answer options data P n and the target user's answered test data T n The analysis question vector X in the optimization correlation calculation is used to obtain the ability growth parameter vector α, α=[α1, α2, …, α i , …, α I ]; The capacity increment calculation layer is used to utilize the formula Z n =αžY Calculate the target user's ability growth factor Z n ; Y is composed of the test data T that the target user has answered n The test question analysis ability vector Y is determined; The capability factor output layer is used to output the target user capability growth factor Z n .

[0007] As a preferred technical solution of the present invention, the specific steps of optimizing the correlation calculation include: For any X i Perform optimized correlation calculation: For the target user's answer option data P n Perform feature extraction to obtain the target user's answer option data feature P n '; According to the target user's answer option data characteristics P n ' and X i Correlation calculation to obtain the data correlation Q of the target user's answer options i ; Using the formula α i =(MM min +β*MM max ) / (Q i +β*MM max ) Calculate α i ;MM min Indicates all Q i The minimum value in MM max Indicates all Q i The maximum value in; β represents the adjustment parameter; Iterate over all X i , we get the capability growth parameter vector α, α=[α1, α2, …, α i , …, α I ].

[0008] As a preferred technical solution of the present invention, the user recommendation test question sentiment analysis model in step S2 includes a duration analysis layer, an interaction data analysis layer and a sentiment fluctuation factor output layer; The duration analysis layer is used to calculate the target user's answer time S n Calculate with the set threshold time to obtain the target user's duration emotion parameter U n ; The interactive data analysis layer is used to analyze the interactive data H of target users’ answers. n Analyze and get the target user emotion prediction label V n ; Predict label V based on target user sentiment n Match in the emotion parameter library to obtain the target user emotion prediction label factor V n '; The emotion fluctuation factor output layer is used to calculate the target user's emotion parameter U according to the target user's duration. n , target user emotion prediction label factor V n 'Calculate the emotion fluctuation factor function to obtain the target user's emotion fluctuation factor E n; The emotion fluctuation factor function is determined by prior data and optimization algorithm; The specific steps for training the interactive data analysis layer include: Collect several groups of emotion label training samples; each group of emotion label training samples contains user answering interaction data and annotated emotion labels; combine several groups of emotion label training samples to obtain an emotion label training set; According to the emotion label training set, the interaction data analysis layer is trained with the goal of labeling emotion labels to obtain an initial interaction data analysis layer; a model evaluation is performed on the initial interaction data analysis layer to obtain an initial interaction data analysis layer model evaluation result; if the initial interaction data analysis layer model evaluation result is passed, the initial interaction data analysis layer is used as the interaction data analysis layer; otherwise, the model training continues.

[0009] As a preferred technical solution of the present invention, the specific steps of determining the emotion fluctuation factor function include: Collect several groups of emotion fluctuation verification samples; in each group of emotion fluctuation verification samples, set the independent variables as user duration emotion parameters and user emotion prediction label factors, and the dependent variable as user emotion fluctuation factors; all emotion fluctuation verification samples are suitable for subsequent analysis; combine several groups of emotion fluctuation verification samples to obtain an emotion fluctuation verification set; Construct K functions to optimize individual G k , k=1, 2, ..., K; each function optimizes individual G k contains a set of solutions for constructing the emotional fluctuation factor function; K functions optimize individual G k Combine to obtain the function optimization iteration population; set the number of iterations j, j = 1, 2, ..., J; set the iteration balance coefficient R t , R t =D*((Jj) / J) 3 , D is the iterative balance weight; During iteration, the function optimization individual G in the function optimization iterative population is k Perform simulation calculation to obtain the function optimization fitness W k ; Determine the iterative balance coefficient R at this time t Is it less than the preset stage update threshold? If the iteration balance coefficient Rt is greater than the preset stage update threshold, the function optimization iteration population is subjected to Gaussian perturbation mutation to obtain a new function optimization iteration population for the next iteration; if the iteration balance coefficient Rt is less than the preset stage update threshold, the function optimization iteration population is subjected to Cauchy perturbation mutation to obtain a new function optimization iteration population for the next iteration; The specific steps for simulation calculation include: Optimize individual G according to the function kCalculate the independent variables in the emotion fluctuation validation set to obtain the simulated emotion fluctuation factor; calculate the Euclidean distance between the simulated emotion fluctuation factor and the user emotion fluctuation factor to obtain the simulated Euclidean distance; use the inverse of the simulated Euclidean distance as a function to optimize the individual G k The function optimizes the fitness W k ; When the number of iterations j reaches the maximum value, the output function optimizes the fitness W k The maximum corresponding function optimizes the individual G k , that is, the optimal function optimization individual; the solution of the corresponding emotional fluctuation factor function constructed in the optimal function optimization individual is taken as the emotional fluctuation factor function.

[0010] As a preferred technical solution of the present invention, the user test question recommendation prediction model in step S3 includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer; The capability increment matching layer is used to increase the target user's capability factor Z. n and target user emotion fluctuation factor E n Calculate and get the target user capability increment Z n '; The comprehensive capability analysis layer is used to calculate the capability increment Z of all target users in the current stage. n 'Perform feature fusion to obtain the target user's stage capability increment; The question recommendation layer is used to recommend questions based on the target user's stage capability increments, and obtain a set of pre-test questions recommended for the target user.

[0011] A personalized test question recommendation system based on a question bank, comprising: A user recommended test question ability analysis module, which includes a user test question acquisition unit and a user test question analysis unit; The user test question acquisition unit is used to acquire the target user's stage answer data; the target user's stage answer data contains N pieces of target user's answered test question data T n and the target user answer option data P n , n=1, 2, …, N; N is the total number of questions answered by the target user in the current stage; The user test question analysis unit is used to analyze the test questions answered by the target user based on the test question data T n , Target user answer option data P n And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; A test question data analysis module, which includes a test question feature acquisition unit and a test question feature analysis unit; The test feature acquisition unit is used to acquire test data for any test question and convert the test data into an analysis test vector X, where X = [X1, X2, ..., X i , …, X I ], X i Represents the specific information of the i-th question feature in the question vector X, where I is the total number of question features in the question data; The question feature analysis unit is used to input the question analysis vector into the personalized question analysis model for analysis, and obtain the question analysis ability vector Y, Y = [Y1, Y2, ..., Y i , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n ; The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; A user-recommended test question sentiment analysis module, which includes a user sentiment acquisition unit and a user sentiment analysis unit; The user emotion acquisition unit is used to obtain the target user's stage answering behavior data. The target user's stage answering behavior data contains the target user's answering time S n Interaction data H with target users n ; The user emotion analysis unit is used to obtain the target user's emotion fluctuation factor E based on the target user's stage answering behavior data and the user recommended test question emotion analysis model. n The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; A user test question recommendation prediction module, which includes a test question recommendation analysis unit; The question recommendation analysis unit is used to increase the ability factor Z of the target user. n , Target user emotion fluctuation factor E nThe user test question recommendation prediction model is analyzed to obtain a target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, corresponding test questions are recommended for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which are used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities.

[0012] The present invention has the following advantages: 1. The present invention can achieve accurate personalized recommendations for users by acquiring the answer data and behavior data of target users and combining them with ability analysis and emotion analysis models. It can recommend the most suitable test questions according to the user's ability growth and emotional fluctuations, thereby improving learning efficiency; by converting the test question data into analysis test question vectors and analyzing them using a personalized test question analysis model, it can comprehensively evaluate the user's mastery of various knowledge points, not only focusing on the user's answer accuracy, but also considering the user's specific performance in the answering process, thereby providing a more comprehensive ability assessment.

[0013] 2. The present invention can effectively identify the user's emotional fluctuations in the process of answering questions by monitoring the user's answering time and interaction data, and combining it with the emotion analysis model, which helps to timely discover the user's learning fatigue or anxiety state, and then dynamically adjust the recommended test questions according to the real-time emotional judgment ability increment and the target user's ability growth factor and emotional fluctuation factor, so that the recommended test questions are more in line with the user's current state, neither too simple nor too difficult, thereby maintaining the user's learning motivation and interest. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The figure is a schematic diagram of the structure of a personalized test question recommendation system based on a question bank adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] Embodiment 1, a personalized test question recommendation method based on a question bank, comprises the following steps: Step S1: Obtain the target stage answer data; the target user stage answer data contains N pieces of target user answered test data T n and the target user answer option data P n , n=1, 2, ..., N; N is the total number of questions answered by the target user in the current stage; based on the data of questions answered by the target user T n , Target user answer option data Pn And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; Get the target user's answered test data T n Specific steps: For any test question, obtain the test question data and convert the test question data into an analysis test question vector X, X=[X1, X2, …, X i , …, X I ], X i represents the specific information of the i-th question feature in the analysis question vector X, where I is the total number of question features in the question data; the analysis question vector is input into the personalized question analysis model for analysis, and the question analysis ability vector Y is obtained, where Y=[Y1, Y2, …, Y i , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; The number and specific content of the test question features are set by professional technicians according to actual conditions; for example, for test question ①, the question stem contains knowledge points I, II, and III; the relationship between each option O1, O2, O3, and O4 and the knowledge point is as follows: O1: correct option, involving knowledge points I, II, and III; O2: incorrect option, involving knowledge points II and III; O3: incorrect option, involving knowledge points I and III; O4: incorrect option, involving knowledge point III; for test question ①, the specific information is the question stem, options, knowledge points, etc., and feature analysis is performed on these specific information to obtain a test question analysis ability vector, which includes knowledge points and ability bases after feature analysis; The personalized test question analysis model in step S1 includes a feature recognition layer, a feature analysis layer C i And feature result output layer; The feature recognition layer is used to analyze all X in the question vector X. i Perform feature recognition and obtain feature label B i ; According to feature label B i X i Send to the corresponding feature analysis layer C i ; Feature Analysis Layer C i For X i Perform feature analysis and obtain Y i ; The feature result output layer is used to convert all Y i Combine them to get the question analysis ability vector Y, Y=[Y1, Y2,…, Y i , …, Y I ]; Training feature analysis layer C i The specific steps include: According to the setting feature label B i Collecting several groups of test question feature training samples corresponding to test question features; each group of test question feature training samples contains a group of specific information corresponding to test question features and test question ability analysis data; combining several groups of test question feature training samples to obtain a test question feature training set; According to the test question feature training set, the feature analysis layer C i Model training is performed with the test ability analysis data as the target to obtain the initial feature analysis layer C i '; Initial feature analysis layer C i 'Evaluate the model and get the initial feature analysis layer C i 'Model evaluation results; if the initial feature analysis layer C i 'If the model evaluation result is passed, the initial feature analysis layer C i 'As feature analysis layer C i , deployed to the personalized test question analysis model, otherwise, continue model training; Traverse and train all feature analysis layers C i , until the model training is completed; By converting the test data into an analysis test vector and using the feature recognition layer for feature recognition, the various features of the test can be accurately analyzed. This method ensures a comprehensive understanding of the test, thereby improving the accuracy and effectiveness of the test analysis; the feature analysis layer conducts a special analysis on each feature to ensure that each feature can be fully processed. This method improves the efficiency of feature analysis and enables the system to quickly analyze a large number of test questions; through the feature recognition layer and the feature analysis layer, personalized ability assessment can be performed based on the specific characteristics of each test question. This method enables the test analysis ability vector to reflect the actual difficulty of the test question and the coverage of knowledge points, thereby providing a reliable foundation for subsequent user ability growth analysis; through accurate test analysis, the most suitable test questions can be recommended to each student, thereby realizing a personalized learning path. This method helps to stimulate students' interest in learning and improve learning efficiency; The user recommended test question ability analysis model in step S1 includes a score judgment layer, an ability increment calculation layer and an ability factor output layer; The scoring judgment layer is used to judge the target user's answer options data P n and the target user's answered test data T n The analysis question vector X in the optimization correlation calculation is used to obtain the ability growth parameter vector α, α=[α1, α2, …, α i , …, α I ]; The capacity increment calculation layer is used to utilize the formula Z n =αžY Calculate the target user's ability growth factor Z n ; Y is composed of the test data T that the target user has answered n The test analysis ability vector Y is determined; The capability factor output layer is used to output the target user capability growth factor Z n ; The specific steps for optimizing correlation calculation include: For any X i Perform optimized correlation calculation: For the target user's answer option data P n Perform feature extraction to obtain the target user's answer option data feature P n '; According to the target user's answer option data characteristics P n ' and X i Correlation calculation to obtain the data correlation Q of the target user's answer options i ; Using the formula α i =(MM min +β*MM max ) / (Q i +β*MM max ) Calculate α i ;MM min Indicates all Q i The minimum value in MM max Indicates all Q i The maximum value in; β represents the adjustment parameter; the specific value of the adjustment parameter is set by professional technicians according to actual conditions; Iterate over all X i , we get the capability growth parameter vector α, α=[α1, α2, …, α i , …, α I ]; By optimizing the correlation calculation, the system can accurately evaluate the target user's ability growth in each test question feature. This method ensures the accuracy and reliability of ability evaluation and provides a solid foundation for subsequent personalized recommendations. By adjusting the parameter β, the ability growth parameter vector α can be adjusted according to the actual situation, which can adapt to the different needs of different users and improve the flexibility and adaptability of recommendations. By extracting the features of the target user's answer option data and calculating its correlation with the test question features, Q i , which can objectively evaluate the user's performance on different features. This method reduces the bias caused by subjective judgment and improves the objectivity of the evaluation; by continuously calculating the ability growth factor Z on each test feature n , can continuously evaluate the user's ability growth, which helps to identify problems in time and take appropriate intervention measures to improve learning effects; Step S2: Obtain the target user's stage answering behavior data, which includes the target user's stay time S for answering questions. n Interaction data H with target users n ; Based on the target user's stage answering behavior data and the user recommended test question sentiment analysis model, the target user's sentiment fluctuation factor E is obtained n The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; The user recommendation test question sentiment analysis model in step S2 includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer; The duration analysis layer is used to calculate the target user's answer time S n Calculate with the set threshold time to obtain the target user's duration emotion parameter U n ; The threshold time is set by the average duration of big data; The interactive data analysis layer is used to analyze the interactive data H of target users’ answers. n Analyze and get the target user emotion prediction label V n ; Predict label V based on target user sentiment n Match in the emotion parameter library to obtain the target user emotion prediction label factor V n '; The emotion parameter library is set based on prior knowledge and contains several standard emotion labels and corresponding influencing factors; The emotion fluctuation factor output layer is used to calculate the target user's emotion parameter U according to the target user's duration. n , target user emotion prediction label factor V n'Calculate the emotion fluctuation factor function to obtain the target user's emotion fluctuation factor E n ; The emotion fluctuation factor function is determined by prior data and optimization algorithm; The specific steps for training the interactive data analysis layer include: Collect several groups of emotion label training samples; each group of emotion label training samples contains user answering interaction data and annotated emotion labels; combine several groups of emotion label training samples to obtain an emotion label training set; According to the emotion label training set, the interaction data analysis layer is trained with the goal of labeling emotion labels to obtain an initial interaction data analysis layer; a model evaluation is performed on the initial interaction data analysis layer to obtain an initial interaction data analysis layer model evaluation result; if the initial interaction data analysis layer model evaluation result is passed, the initial interaction data analysis layer is used as the interaction data analysis layer; otherwise, the model training continues; By analyzing the target user's answering time and answering interaction data, the user's emotional fluctuations can be accurately monitored. This method ensures real-time monitoring of the user's emotional state, which helps to promptly discover and solve the user's emotional problems in the learning process; through the comprehensive evaluation of the time analysis layer and the interaction data analysis layer, the user's emotional state can be comprehensively analyzed from multiple dimensions. This method improves the comprehensiveness and accuracy of emotional evaluation; by analyzing the user's answering interaction data and matching it with emotional tags, the user's emotional state can be accurately predicted. This method improves the accuracy of emotional prediction and helps to better understand the user's behavioral motivations; through the calculation of the emotional fluctuation factor function, the user's emotional fluctuations can be quantified, which is convenient for subsequent analysis and application. This method makes the evaluation of emotional fluctuations more objective and quantitative; The specific steps to determine the emotional fluctuation factor function include: Collect several groups of emotion fluctuation verification samples; in each group of emotion fluctuation verification samples, set the independent variables as user duration emotion parameters and user emotion prediction label factors, and the dependent variable as user emotion fluctuation factors; all emotion fluctuation verification samples are suitable for subsequent analysis; combine several groups of emotion fluctuation verification samples to obtain an emotion fluctuation verification set; Construct K functions to optimize individual G k , k=1, 2, ..., K; each function optimizes individual G k contains a set of solutions for constructing the emotional fluctuation factor function; K functions optimize individual G k Combine to obtain the function optimization iteration population; set the number of iterations j, j = 1, 2, ..., J; set the iteration balance coefficient R t , R t =D*((Jj) / J) 3 , D is the iterative balance weight; The iterative balance coefficient changes quickly in the early stage of iteration, which can improve the global search ability of the optimal individual and prevent it from falling into the local optimum. In the later stage of iteration, the iterative balance coefficient changes slowly, which is conducive to the function optimization individual to explore the optimal solution and improve the local search ability and search accuracy of the algorithm. During iteration, the function optimization individual G in the function optimization iterative population is k Perform simulation calculation to obtain the function optimization fitness W k ; Determine the iterative balance coefficient R at this time t Is it less than the preset stage update threshold? If the iteration balance coefficient Rt is greater than the preset stage update threshold, the function optimization iteration population is subjected to Gaussian perturbation mutation to obtain a new function optimization iteration population for the next iteration; if the iteration balance coefficient Rt is less than the preset stage update threshold, the function optimization iteration population is subjected to Cauchy perturbation mutation to obtain a new function optimization iteration population for the next iteration; the preset stage update threshold is set by professional and technical personnel according to actual conditions; The optimization algorithm conducts a large-scale search in the early iteration stage to gradually narrow the target position range, and conducts a small-scale search in the later iteration stage to find the precise position of the optimal individual. As the population iterates, some inferior solutions will appear and cannot be eliminated, which can easily cause the algorithm to fall into the local optimum. Therefore, two different mutation mechanisms are introduced to help the iterative population jump out of the local optimum. The specific steps for simulation calculation include: Optimize individual G according to the function k Calculate the independent variables in the emotion fluctuation validation set to obtain the simulated emotion fluctuation factor; calculate the Euclidean distance between the simulated emotion fluctuation factor and the user emotion fluctuation factor to obtain the simulated Euclidean distance; use the inverse of the simulated Euclidean distance as a function to optimize the individual G k The function optimizes the fitness W k ; When the number of iterations j reaches the maximum value, the output function optimizes the fitness W k The maximum corresponding function optimizes the individual G k , that is, the optimal function optimization individual; the solution of the corresponding emotional fluctuation factor function constructed in the optimal function optimization individual is used as the emotional fluctuation factor function; By collecting emotion fluctuation verification samples and performing iterative calculations of the optimization algorithm, the system can determine the most appropriate function to describe the user's emotion fluctuation factor, improve the accuracy of the emotion fluctuation factor function, and make the evaluation of emotion fluctuation more accurate; by setting the iterative balance coefficient and using Gaussian perturbation mutation in the early stage of iteration and Cauchy perturbation mutation in the later stage, local search can be gradually performed on the basis of global search to avoid falling into the local optimal solution. This method improves the algorithm's search ability and search accuracy; the dynamic adjustment of the iterative balance coefficient enables the algorithm to adopt different search strategies at different stages; in the early stage, the algorithm conducts a large-scale search and gradually narrows the target position range; in the later stage, the algorithm conducts a small-scale search to find the precise position of the optimal individual. This method enables the algorithm to adapt to the needs of different stages more flexibly; by calculating the independent variables in the emotion fluctuation verification set and comparing them with the actual emotion fluctuation factors, the system can optimize based on data. This method makes the optimization process more objective and reliable, and reduces the deviation caused by subjective judgment; Step S3: According to the target user's ability growth factor Z n , Target user emotion fluctuation factor E n Analyze with the user test question recommendation prediction model to obtain the target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, recommend corresponding test questions for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which is used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities; The user test question recommendation prediction model in step S3 includes an ability increment matching layer, a comprehensive ability analysis layer, and a test question recommendation layer; The capability increment matching layer is used to increase the target user's capability factor Z. n and target user emotion fluctuation factor E n Calculate and get the target user capability increment Z n '; The comprehensive capability analysis layer is used to calculate the capability increment Z of all target users in the current stage. n 'Perform feature fusion to obtain the target user's stage capability increment; The test question recommendation layer is used to recommend test questions based on the target user's stage capability increment, and obtain the target user's test question recommendation pre-test question set; By comprehensively considering the target user's ability growth factor and emotional fluctuation factor, the most suitable test questions can be recommended to each user. This method ensures that the recommended test questions are consistent with the user's current ability level and take into account the user's emotional state, thereby improving the learning effect; through the ability increment matching layer and the comprehensive ability analysis layer, the user's ability growth can be comprehensively analyzed. This method not only focuses on the user's correct answer rate, but also considers the user's specific performance in the answering process, thereby providing a more comprehensive ability analysis; according to the user's ability increment Z n ′ and emotional fluctuation factor E n , can dynamically adjust the recommended test questions. This method makes the recommended test questions more in line with the user's current status, neither too simple nor too difficult, thereby maintaining the user's learning motivation.

[0017] Example 2, a personalized test question recommendation system based on a question bank, see Figure 1 As shown, including: A user recommended test question ability analysis module, which includes a user test question acquisition unit and a user test question analysis unit; The user test question acquisition unit is used to acquire the target user's stage answer data; the target user's stage answer data contains N pieces of target user's answered test question data T n and the target user answer option data P n , n=1, 2, …, N; N is the total number of questions answered by the target user in the current stage; The user test question analysis unit is used to analyze the test questions answered by the target user based on the test question data T n , Target user answer option data P n And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; A test question data analysis module, which includes a test question feature acquisition unit and a test question feature analysis unit; The test feature acquisition unit is used to acquire test data for any test question and convert the test data into an analysis test vector X, where X = [X1, X2, ..., X i , …, X I ], X i Represents the specific information of the i-th question feature in the question vector X, where I is the total number of question features in the question data; The question feature analysis unit is used to input the question analysis vector into the personalized question analysis model for analysis, and obtain the question analysis ability vector Y, Y = [Y1, Y2, ..., Yi , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; A user-recommended test question sentiment analysis module, which includes a user sentiment acquisition unit and a user sentiment analysis unit; The user emotion acquisition unit is used to obtain the target user's stage answering behavior data. The target user's stage answering behavior data contains the target user's answering time S n Interaction data H with target users n ; The user emotion analysis unit is used to obtain the target user's emotion fluctuation factor E based on the target user's stage answering behavior data and the user recommended test question emotion analysis model. n The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; A user test question recommendation prediction module, which includes a test question recommendation analysis unit; The question recommendation analysis unit is used to increase the ability factor Z of the target user. n , Target user emotion fluctuation factor E n The user test question recommendation prediction model is analyzed to obtain a target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, corresponding test questions are recommended for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which are used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities.

[0018] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A personalized test question recommendation method based on a question bank, characterized in that: The following steps are involved: Step S1: Obtaining the target user's stage answer data; The target user stage answer data contains N pieces of target user answered test data T n and the target user answer option data P n , n=1,2,…,N; N is the total number of test questions that the target user has answered in the current stage; Based on the target user's answered test data T n , Target user answer option data P n And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; Get the target user's answered test data T n Specific steps: For any test question, obtain the test question data and convert the test question data into an analysis test question vector X, X=[X1, X2, …, X i , …, X I ], X i Represents the specific information of the i-th question feature in the question vector X, where I is the total number of question features in the question data; Input the analysis question vector into the personalized question analysis model for analysis, and obtain the question analysis ability vector Y, Y = [Y1, Y2, ..., Y i , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; Step S2: Obtain the target user's stage answering behavior data, which includes the target user's stay time S for answering questions. n Interaction data H with target users n ; Based on the target user's stage answering behavior data and the user recommended test question sentiment analysis model, the target user's sentiment fluctuation factor E is obtained n ; The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; Step S3: According to the target user capability growth factor Z n , Target user emotion fluctuation factor E n The user test question recommendation prediction model is analyzed to obtain a target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, corresponding test questions are recommended for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which are used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities.

2. A personalized test question recommendation method based on a question bank according to claim 1, characterized in that: The personalized test question analysis model in step S1 includes a feature recognition layer, a feature analysis layer C i And feature result output layer; The feature recognition layer is used to analyze all X in the question vector X. i Perform feature recognition and obtain feature label B i ; According to feature label B i X i Send to the corresponding feature analysis layer C i ; Feature Analysis Layer C i For X i Perform feature analysis and obtain Y i ; The feature result output layer is used to convert all Y i Combine them to get the question analysis ability vector Y, Y=[Y1, Y2,…, Y i , …, Y I ]; Training feature analysis layer C i The specific steps include: According to the setting feature label B i Collecting several groups of test question feature training samples corresponding to test question features; each group of test question feature training samples contains a group of specific information corresponding to test question features and test question ability analysis data; combining several groups of test question feature training samples to obtain a test question feature training set; According to the test feature training set, the feature analysis layer C i Model training is performed with the test ability analysis data as the target to obtain the initial feature analysis layer C i '; Initial feature analysis layer C i 'Evaluate the model and get the initial feature analysis layer C i 'Model evaluation results; if the initial feature analysis layer C i 'If the model evaluation result is passed, the initial feature analysis layer C i 'As feature analysis layer C i , deployed to the personalized test question analysis model, otherwise, continue model training; Traverse and train all feature analysis layers C i , until the model training is completed.

3. A personalized test question recommendation method based on a question bank according to claim 2, characterized in that: The user recommended test question ability analysis model in step S1 includes a score judgment layer, an ability increment calculation layer and an ability factor output layer; The scoring judgment layer is used to judge the target user's answer options data P n and the target user's answered test data T n The analysis question vector X in the optimization correlation calculation is used to obtain the ability growth parameter vector α, α=[α1, α2, …, α i , …, α I ]; The capacity increment calculation layer is used to utilize the formula Z n =αžY Calculate the target user's ability growth factor Z n ; Y is composed of the test data T that the target user has answered n The test analysis ability vector Y is determined; The capability factor output layer is used to output the target user capability growth factor Z n .

4. A personalized test question recommendation method based on a question bank according to claim 3, characterized in that: The specific steps for optimizing correlation calculation include: For any X i Perform optimized correlation calculation: For the target user's answer option data P n Perform feature extraction to obtain the target user's answer option data feature P n '; According to the target user's answer option data characteristics P n ' and X i Correlation calculation to obtain the data correlation Q of the target user's answer options i ; Using the formula α i =(MM min +β*MM max ) / (Q i +β*MM max ) Calculate α i ;MM min Indicates all Q i The minimum value in MM max Indicates all Q i The maximum value in; β represents the adjustment parameter; Iterate over all X i , we get the capability growth parameter vector α, α=[α1, α2, …, α i , …, α I ].

5. A personalized test question recommendation method based on a question bank according to claim 4, characterized in that: The user recommendation test question sentiment analysis model in step S2 includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer; The duration analysis layer is used to calculate the target user's answer time S n Calculate with the set threshold time to obtain the target user's duration emotion parameter U n ; The interactive data analysis layer is used to analyze the interactive data H of target users’ answers. n Analyze and get the target user emotion prediction label V n ; Predict label V based on target user sentiment n Match in the emotion parameter library to obtain the target user emotion prediction label factor V n '; The emotion fluctuation factor output layer is used to calculate the target user's emotion parameter U according to the target user's duration. n , target user emotion prediction label factor V n 'Calculate the emotion fluctuation factor function to obtain the target user's emotion fluctuation factor E n ; The emotional fluctuation factor function is determined by prior data and optimization algorithms; The specific steps for training the interactive data analysis layer include: Collect several groups of emotion label training samples; each group of emotion label training samples contains user answering interaction data and annotated emotion labels; combine several groups of emotion label training samples to obtain an emotion label training set; According to the emotion label training set, the interaction data analysis layer is trained with the goal of labeling emotion labels to obtain an initial interaction data analysis layer; a model evaluation is performed on the initial interaction data analysis layer to obtain an initial interaction data analysis layer model evaluation result; if the initial interaction data analysis layer model evaluation result is passed, the initial interaction data analysis layer is used as the interaction data analysis layer; otherwise, the model training continues.

6. A personalized test question recommendation method based on a question bank according to claim 5, characterized in that: The specific steps to determine the emotional fluctuation factor function include: Collect several groups of emotion fluctuation verification samples; in each group of emotion fluctuation verification samples, set the independent variables as user duration emotion parameters and user emotion prediction label factors, and the dependent variable as user emotion fluctuation factors; all emotion fluctuation verification samples are suitable for subsequent analysis; combine several groups of emotion fluctuation verification samples to obtain an emotion fluctuation verification set; Construct K functions to optimize individual G k , k=1, 2, ..., K; each function optimizes individual G k contains a set of solutions for constructing the emotional fluctuation factor function; K functions optimize individual G k Combine to obtain the function optimization iteration population; set the number of iterations j, j = 1, 2, ..., J; set the iteration balance coefficient R t , R t =D*((Jj) / J) 3 , D is the iterative balance weight; During iteration, the function optimization individual G in the function optimization iterative population is k Perform simulation calculation to obtain the function optimization fitness W k ; Determine the iterative balance coefficient R at this time t Is it less than the preset stage update threshold? If the iteration balance coefficient Rt is greater than the preset stage update threshold, the function optimization iteration population is subjected to Gaussian perturbation mutation to obtain a new function optimization iteration population for the next iteration; if the iteration balance coefficient Rt is less than the preset stage update threshold, the function optimization iteration population is subjected to Cauchy perturbation mutation to obtain a new function optimization iteration population for the next iteration; The specific steps for simulation calculation include: Optimize individual G according to the function k Calculate the independent variables in the emotion fluctuation validation set to obtain the simulated emotion fluctuation factor; calculate the Euclidean distance between the simulated emotion fluctuation factor and the user emotion fluctuation factor to obtain the simulated Euclidean distance; use the inverse of the simulated Euclidean distance as a function to optimize the individual G k The function optimizes the fitness W k ; When the number of iterations j reaches the maximum value, the output function optimizes the fitness W k The maximum corresponding function optimizes the individual G k , that is, the optimal function optimization individual; the solution of the corresponding emotional fluctuation factor function constructed in the optimal function optimization individual is taken as the emotional fluctuation factor function.

7. A personalized test question recommendation method based on a question bank according to claim 6, characterized in that: The user test question recommendation prediction model in step S3 includes an ability increment matching layer, a comprehensive ability analysis layer, and a test question recommendation layer; The capability increment matching layer is used to increase the target user's capability factor Z. n and target user emotion fluctuation factor E n Calculate and get the target user capability increment Z n '; The comprehensive capability analysis layer is used to calculate the capability increment Z of all target users in the current stage. n 'Perform feature fusion to obtain the target user's stage capability increment; The question recommendation layer is used to recommend questions based on the target user's stage capability increments, and obtain a set of pre-test questions recommended for the target user.

8. A personalized test question recommendation system based on a question bank, characterized in that: The system applies a personalized test question recommendation method based on a question bank as described in any one of claims 1 to 7, including: A user recommended test question ability analysis module, which includes a user test question acquisition unit and a user test question analysis unit; The user test question acquisition unit is used to acquire the target user's stage answer data; the target user's stage answer data contains N pieces of target user's answered test question data T n and the target user answer option data P n , n=1, 2, …, N; N is the total number of questions answered by the target user in the current stage; The user test question analysis unit is used to analyze the test questions answered by the target user based on the test question data T n , Target user answer option data P n And the user recommendation test question ability analysis model, get the target user ability growth factor Z n The user recommendation test question ability analysis model includes a score judgment layer, an ability increment calculation layer, and an ability factor output layer, which is used to analyze the target user's ability growth based on the specific characteristics of the answered test questions; A test question data analysis module, which includes a test question feature acquisition unit and a test question feature analysis unit; The test feature acquisition unit is used to acquire test data for any test question and convert the test data into an analysis test vector X, where X = [X1, X2, ..., X i , …, X I ], X i Represents the specific information of the i-th question feature in the question vector X, where I is the total number of question features in the question data; The question feature analysis unit is used to input the question analysis vector into the personalized question analysis model for analysis, and obtain the question analysis ability vector Y, Y = [Y1, Y2, ..., Y i , …, Y I] , Y i For X i The question ability analysis data corresponding to the i-th question feature; the analysis question vector X and the question analysis ability vector Y corresponding to the question answered by the target user are used as the question data T of the target user answered n The personalized test question analysis model includes a feature recognition layer, a feature analysis layer C i and the feature result output layer, which is used to conduct specific and accurate feature analysis on the test questions; A user-recommended test question sentiment analysis module, which includes a user sentiment acquisition unit and a user sentiment analysis unit; The user emotion acquisition unit is used to obtain the target user's stage answering behavior data. The target user's stage answering behavior data contains the target user's answering time S n Interaction data H with target users n ; The user emotion analysis unit is used to obtain the target user's emotion fluctuation factor E based on the target user's stage answering behavior data and the user recommended test question emotion analysis model. n The user recommendation test question sentiment analysis model includes a duration analysis layer, an interaction data analysis layer, and a sentiment fluctuation factor output layer, which is used to quantify the user's sentiment when answering the test questions using the sentiment fluctuation factor function, and determine the sentiment fluctuation factor function using the improved optimization algorithm; A user test question recommendation prediction module, which includes a test question recommendation analysis unit; The question recommendation analysis unit is used to increase the ability factor Z of the target user. n , Target user emotion fluctuation factor E n The user test question recommendation prediction model is analyzed to obtain a target user test question recommendation pre-test question set; based on the target user test question recommendation pre-test question set, corresponding test questions are recommended for the target user in the question bank in the next stage; the user test question recommendation prediction model includes an ability increment matching layer, a comprehensive ability analysis layer and a test question recommendation layer, which are used to perform comprehensive ability analysis on the target user and match personalized test questions with corresponding abilities.

Images