CN-122019888-A - Question recommendation method and system based on user learning portrait analysis

CN122019888ACN 122019888 ACN122019888 ACN 122019888ACN-122019888-A

Abstract

The invention discloses a topic recommendation method and a system based on user learning portraits analysis, which belong to the technical field of intelligent education, and are used for acquiring initial learning portraits characteristic vectors of target users in a learning period, constructing a user portraits evolution matrix set, quantifying a mastery change abnormality index, an interest offset and a time distribution offset value, constructing a staged user portraits evolution function model based on the initial portraits and the evolution matrix and used for simulating a dynamic migration process of a learning state, identifying a knowledge blank interval and a potential concept confusion section according to the predicted user portraits state and generating target learning task vectors, and finally executing screening and matching in a topic library according to the learning task vectors to output personalized recommendation topic sets.

Inventors

WANG CHUANJUN
XU FEI
HU KONGWANG

Assignees

上海贝乘教育科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260226

Claims (9)

1. A topic recommendation method based on user learning portrait analysis is characterized by comprising the following steps: s100, obtaining initial learning portrait feature vectors of a target user in a learning period, wherein the initial learning portrait feature vectors comprise knowledge point mastering degree, question making accuracy, operation duration, learning time distribution and wrong question concentration; S200, acquiring a behavior data sequence corresponding to each learning stage of a target user in a learning period, and constructing a user portrait evolution matrix group A= { A1, A2, ai, an } according to the behavior data change, wherein each Ai represents portrait change characteristics of the user in An i-th learning stage, an is the maximum period state of portrait change of the user, and Ai comprises a grasping degree change abnormality index, interest offset and time distribution offset value; S300, constructing a staged user portraits evolution function model F (x) based on a user portraits evolution matrix group A and an initial learning portraits feature vector, wherein x is a time step and is used for simulating a dynamic migration process of a user learning state; S400, according to the model F (x), combining the user portrait state Ui under the current time step, calculating a knowledge blank section and a potential confusion section of the user under the portrait state Ui, and generating a target learning task vector; s500, based on the target learning task vector, performing topic screening and precision matching in a topic library, generating a corresponding recommended topic set, and pushing the recommended topic set to a target user.
2. The method for question recommendation based on user learning portrayal analysis according to claim 1, wherein the calculation of the grasping level change abnormality index comprises: extracting knowledge point grasping degree sequences of each stage of a user according to a learning stage sequence and constructing continuous time sequence data to serve as variable point detection input data; Performing global segmentation optimization calculation on the time sequence by using a pruning accurate linear time algorithm based on variable point detection, identifying a structure mutation position in the mastering degree change process and outputting a stage variable point set; calculating corresponding variation amplitude according to the average difference of the mastery degrees of the front stage and the rear stage of each variation point, and forming a stage variation intensity sequence by combining the occurrence frequency of the variation point; And carrying out normalization processing on the stage change intensity sequence and generating a mastery degree change abnormality index which is used for representing the abnormality degree of mastery degree fluctuation in the learning state of the user.
3. The method for question recommendation based on user learning portrayal analysis of claim 1, wherein the calculating of the interest offset comprises: according to the question answer records and the knowledge point access frequency of the user in each learning stage, extracting the interest topic distribution of each stage, determining the interest state nodes of the corresponding stage according to the knowledge topic with the highest occurrence frequency, and constructing an interest state transition path according to the learning stage sequence; counting the state transition times based on the transition relation of the interest states between adjacent learning stages, calculating the transition probability between the interest states, constructing an interest state transition probability matrix, and further generating a Markov interest state transition diagram; calculating uncertainty change quantity of interest state migration at each stage according to the interest state transition probability matrix, extracting migration probability mutation positions and forming an interest change strength sequence; and carrying out normalization processing on the interest change intensity sequence to generate an interest offset degree which is used for representing the offset degree of the interest direction of the user in the learning process.
4. The method for topic recommendation based on user learning portrayal analysis of claim 1, wherein constructing the staged user portrayal evolution function model F (x) comprises: Combining the portrait change feature vectors of each stage in the user portrait evolution matrix group with the initial learning portrait feature vectors according to the time sequence to form a multidimensional time sequence training sample; Constructing a recurrent neural network structure based on training samples, taking a time step as an input sequence index, and extracting long-term dependence characteristics of user portraits changing along with time by adopting a double-layer gating circulating unit; Training iteration is carried out by taking the mean square error as a loss function, and weight parameters are optimized to fit the dynamic evolution trend of the user image; And defining the model function after training as a staged user portrayal evolution function model, taking the time step as input, and outputting a user portrayal prediction result at the current moment.
5. The method for question recommendation based on user learning portrayal analysis according to claim 1, wherein the step of calculating a knowledge blank section and a potential confusion section of the user in the portrayal state Ui comprises the following steps: Based on the staged user portrait evolution function model, taking the current time step as input to acquire a user portrait state vector under the corresponding time step; Extracting knowledge point mastering degree and wrong question concentration degree indexes in the user portrait state, constructing a knowledge point mastering degree scoring table, and setting a mastering degree threshold value to identify a knowledge vacancy zone; locating a potential concept confusion section by analyzing the difference of mastery degrees and the error distribution density of users on similar knowledge points; And carrying out structural coding on the identified knowledge blank interval, the confusion section and the corresponding question type preference to generate a target learning task vector.
6. The method for question recommendation based on user learning portrayal analysis of claim 5, wherein the identifying knowledge blank section comprises the steps of: extracting grasping degree indexes and wrong question concentration degree indexes corresponding to all knowledge points from the user portrait state vector under the current time step; constructing a knowledge point grasping degree scoring table according to the knowledge point numbers, wherein the scoring value is calculated by the weighted combination of grasping degree of the knowledge point and error question concentration degree; setting a knowledge point mastering degree identification threshold value, and marking the knowledge points with scores lower than the threshold value as contents to be enhanced; the continuous or associated low scoring knowledge points are combined to form a knowledge gap interval.
7. The method for topic recommendation based on user learning portrayal analysis of claim 5, wherein locating potential concept confusion segments comprises the steps of: Constructing a knowledge point similarity map, and calculating a similarity score of each knowledge point pair based on the teaching content relevance and the precedence dependence among the knowledge points; Combining knowledge point mastery degree scores in the current user portrait state, and performing mastery degree difference calculation on all high-similarity knowledge point pairs; screening out knowledge point pairs with the grasping degree difference exceeding a set deviation threshold, and counting the number of errors of the knowledge point pairs in a user history answer record; And calculating confusion risk scores by combining the mastery difference values and the error distribution densities, wherein knowledge point pairs with scores higher than a confusion judgment threshold value are judged to be potential concept confusion sections.
8. The method for question recommendation based on user learning portrayal analysis of claim 1, wherein the generating the corresponding recommendation question set comprises the steps of: Based on the knowledge point numbers, the question type preferences and the difficulty range in the target learning task vector, a primary question set meeting the conditions is screened from a preset question library; Extracting knowledge point coverage characteristics, question labels and historical answer expression labels of each initial choice question, and constructing a question feature vector; Carrying out multidimensional feature matching on the topic feature vector and the current user portrait state vector, and evaluating the matching degree of each topic and the current learning state of the user by adopting a vector included angle cosine similarity calculation mode; And sorting in descending order according to the matching degree score, and selecting the questions with the matching degree higher than a set threshold value to form a recommendation question set as personalized learning content pushed to the user in the current time step.
9. A topic recommendation system based on user learning portrayal analysis, for implementing the topic recommendation method based on user learning portrayal analysis according to any one of claims 1-8, characterized in that it comprises: The user portrait initializing module is used for acquiring initial learning portrait feature vectors of a target user in a learning period, wherein the initial learning portrait feature vectors comprise knowledge point mastering degree, question making accuracy, operation time, learning time distribution and wrong question concentration; The user portrait evolution modeling module acquires a behavior data sequence corresponding to each learning stage of a target user in a learning period, and constructs a user portrait evolution matrix group A= { A1, A2, & gt, ai, & gt, an }, wherein each Ai represents portrait change characteristics of the user in An i-th learning stage, an is the maximum period state of the user portrait change, and Ai comprises a grasping degree change abnormality index, an interest offset degree and a time distribution offset value; the user portrayal prediction modeling module is used for constructing a staged user portrayal evolution function model F (x) based on the user portrayal evolution matrix group A and the initial learning portrayal feature vector, wherein x is a time step and is used for simulating a dynamic migration process of a user learning state; the learning task generating module is used for calculating a knowledge blank section and a potential confusion section of a user in the portrait state Ui according to the model F (x) and combining the portrait state Ui of the user in the current time step to generate a target learning task vector; and the topic matching and recommending module is used for executing topic screening and precision matching in the topic library based on the target learning task vector, generating a corresponding recommending topic set and pushing the recommending topic set to the target user.

Description

Question recommendation method and system based on user learning portrait analysis Technical Field The invention relates to the technical field of intelligent education, in particular to a question recommending method and a system based on user learning portrait analysis. Background Under the background of rapid development of digital education and online learning platforms, the topic recommendation system has become an important tool for improving learning efficiency and personalized teaching experience. The existing topic recommendation method is mostly based on collaborative filtering, knowledge graph or deep learning and other algorithms for modeling, but has obvious defects in the practical application process. Especially, under the situation that learning states of learners frequently change in short time and user interest portrait blurring caused by platform interaction data sparseness, the current real learning requirement of the user is difficult to accurately capture by the existing method, and phenomena such as repeated recommended content, study key separation, unmatched difficulty and the like are easy to occur, so that learning enthusiasm and continuous use experience of the user are affected. For example, when taking a test with high intensity, short period and large content span, the learning level and learning behavior of the learner can fluctuate drastically, and the learning portrait shows high dynamic and staged characteristics. When the conventional static modeling topic recommendation system is used for coping with such dynamic learning paths, the recommendation result is delayed from the user requirement due to the fact that the user portrait evolution process cannot be effectively simulated, and the learning effect is seriously affected. In addition, the lack of a system for modeling the depth of user portrait change makes it difficult to quantitatively identify the "learning fatigue points" or "concept confusion zones" of users, which is particularly prominent in a high-pressure short-time training environment, and is extremely easy to cause knowledge error zone accumulation and learning progress breakage. Therefore, a method capable of accurately simulating microscopic changes of user portraits along with the learning process and dynamically adjusting topic recommendation strategies according to the microscopic changes is needed, so as to solve core technical problems such as recommendation failure, user portraits distortion, learning path breakage and the like, and becomes one of key problems to be solved in the current intelligent education field. Disclosure of Invention The invention aims to provide a question recommending method and a system based on user learning portrait analysis, which are used for solving the defects in the background technology. In order to achieve the above purpose, the invention provides a method for recommending topics based on user learning portrait analysis, which comprises the following steps: s100, obtaining initial learning portrait feature vectors of a target user in a learning period, wherein the initial learning portrait feature vectors comprise knowledge point mastering degree, question making accuracy, operation duration, learning time distribution and wrong question concentration; S200, acquiring a behavior data sequence corresponding to each learning stage of a target user in a learning period, and constructing a user portrait evolution matrix group A= { A1, A2, ai, an } according to the behavior data change, wherein each Ai represents portrait change characteristics of the user in An i-th learning stage, an is the maximum period state of portrait change of the user, and Ai comprises a grasping degree change abnormality index, interest offset and time distribution offset value; S300, constructing a staged user portraits evolution function model F (x) based on a user portraits evolution matrix group A and an initial learning portraits feature vector, wherein x is a time step and is used for simulating a dynamic migration process of a user learning state; S400, according to the model F (x), combining the user portrait state Ui under the current time step, calculating a knowledge blank section and a potential confusion section of the user under the portrait state Ui, and generating a target learning task vector; s500, based on the target learning task vector, performing topic screening and precision matching in a topic library, generating a corresponding recommended topic set, and pushing the recommended topic set to a target user. Preferably, the calculation of the grasping degree change abnormality index includes: extracting knowledge point grasping degree sequences of each stage of a user according to a learning stage sequence and constructing continuous time sequence data to serve as variable point detection input data; Performing global segmentation optimization calculation on the time sequence by using a pruning accurate linear time algor