Knowledge point recommendation method and device, storage medium and electronic device

Abstract

This application provides a knowledge point recommendation method, apparatus, storage medium, and electronic device, relating to the field of online learning technology. The knowledge point recommendation method includes: identifying the target user's weak knowledge points among multiple knowledge points; determining the target user's learning vectors for the multiple knowledge points, whereby the learning vectors represent the target user's mastery of the multiple knowledge points; and recommending superior knowledge points to the target user based on the target user's weak knowledge points and the target user's learning vectors for the multiple knowledge points. In this application, knowledge point recommendations are tailored to the target user's learning ability, which can improve the target user's learning benefits and thus enhance the target user's learning confidence.

Description

Technical Field

[0001] This application relates to the field of online learning technology, specifically to a knowledge point recommendation method, apparatus, storage medium, and electronic device. Background Technology

[0002] Currently, online learning is becoming increasingly popular, and many personalized recommendation methods have emerged in the market. However, can these methods truly achieve personalized diagnosis, personalized recommendations, and effective learning? In other words, how can they accurately answer what users should learn, how they should learn, and how effectively they should learn based on their current learning level? Summary of the Invention

[0003] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method, apparatus, storage medium, and electronic device for recommending knowledge points.

[0004] In a first aspect, one embodiment of this application provides a knowledge point recommendation method, including: determining the weak knowledge points of a target user among multiple knowledge points; determining the learning vector of the target user for multiple knowledge points, wherein the learning vector represents the target user's mastery of multiple knowledge points; and recommending superior knowledge points to the target user based on the target user's weak knowledge points among multiple knowledge points and the target user's learning vector for multiple knowledge points.

[0005] In conjunction with the first aspect, in certain implementations of the first aspect, based on the target user's weak knowledge points across multiple knowledge points and the target user's learning vectors for those multiple knowledge points, recommended superior knowledge points are made to the target user. This includes: determining the knowledge point discrimination degree corresponding to the weak knowledge points, whereby the knowledge point discrimination degree represents the historical mastery information of users with similar knowledge points to the target user regarding the weak knowledge points; determining the knowledge point weakness degree corresponding to the weak knowledge points based on the target user's learning vectors for those multiple knowledge points, whereby the knowledge point weakness degree represents the target user's future mastery information regarding the weak knowledge points; and recommending superior knowledge points to the target user based on the knowledge point discrimination degree and the knowledge point weakness degree corresponding to the weak knowledge points.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, based on the knowledge point differentiation and knowledge point weakness corresponding to weak knowledge points, superior learning knowledge points are recommended to the target user. This includes: obtaining the difference data between the knowledge point differentiation and knowledge point weakness corresponding to weak knowledge points; determining the weight of the knowledge point differentiation corresponding to weak knowledge points based on the difference data; determining the weight of the knowledge point weakness corresponding to weak knowledge points based on the difference data; and recommending superior learning knowledge points to the target user from among the weak knowledge points based on the weight of the knowledge point differentiation and the weight of the knowledge point weakness.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the knowledge point discrimination degree corresponding to the weak knowledge points is determined based on the weak knowledge points, including: determining the same-score and isomorphic users corresponding to the target user based on the learning vectors of the target user for multiple knowledge points; determining the improved users and regressed users among the same-score and isomorphic users based on the learning information corresponding to the same-score and isomorphic users; and determining the knowledge point discrimination degree corresponding to the weak knowledge points based on the score information of improved users on weak knowledge points and the score information of regressed users on weak knowledge points.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the knowledge point weakness corresponding to the weak knowledge point is determined based on the learning vectors of the target user for multiple knowledge points. This includes: determining the feature vectors corresponding to each of the multiple knowledge points, where the feature vectors corresponding to the knowledge points represent the user's average mastery of the knowledge points; and using a score prediction model based on the learning vectors of the target user for multiple knowledge points and the feature vectors corresponding to each of the multiple knowledge points to determine the knowledge point weakness corresponding to the weak knowledge point.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the training method of the score prediction model includes: acquiring several sets of first training datasets, each set of first training datasets including sample questions and the masked question scores or unmasked question scores corresponding to the sample questions; and training the first model to be trained based on the several sets of first training datasets to obtain the score prediction model.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, a first model to be trained is trained based on several sets of first training datasets to obtain a score prediction model, including: sorting several sets of first training datasets based on the answering time of sample questions in the first training datasets to obtain a training dataset sequence corresponding to several sets of first training datasets; and training the first model to be trained based on the training dataset sequence to obtain a score prediction model.

[0011] In conjunction with the first aspect, in certain implementations of the first aspect, determining the target user's weak knowledge points across multiple knowledge areas includes: obtaining the target user's exam scores within a preset time period; determining the exam scores of reference users under similar exam conditions based on the target user's exam scores within the preset time period; identifying users with the same scores as the target user from among the reference users based on the reference users' exam scores and the target user's exam scores; and determining the target user's weak knowledge points across multiple knowledge areas based on the historical exam information of the users with the same scores.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, based on the historical exam information of users with the same score, the weak knowledge points of the target user in multiple knowledge points are determined, including: based on the historical exam information of users with the same score, determining the learning vectors of users with the same score for multiple knowledge points; based on the similarity between the learning vectors of users with the same score for multiple knowledge points and the learning vectors of the target user for multiple knowledge points, determining the users with the same score and the same structure as the target user; and based on the historical exam information of users with the same score and the same structure as the target user, determining the weak knowledge points of the target user in multiple knowledge points.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, based on the historical exam information of users with similar scores and structures, the weak knowledge points of the target user in multiple knowledge points are determined, including: based on the historical exam information of users with similar scores and structures, determining the mastery of users with similar scores and structures in multiple knowledge points; for each knowledge point in multiple knowledge points, if there are a preset number of users with similar scores and structures whose mastery of the knowledge point meets the weak knowledge point setting conditions, then the knowledge point is determined as the weak knowledge point of the target user.

[0014] In conjunction with the first aspect, in certain implementations of the first aspect, determining the learning vectors for multiple knowledge points for the target user includes: obtaining the target user's historical test questions; determining the feature vectors corresponding to each of the historical test questions, wherein the feature vectors represent the target user's answering time, score information and average score rate, difficulty information of the historical test questions and the knowledge points corresponding to the historical test questions; and determining the learning vectors for multiple knowledge points for the target user based on the feature vectors corresponding to each of the historical test questions.

[0015] In conjunction with the first aspect, in some implementations of the first aspect, determining the feature vectors corresponding to each historically answered question includes: using a question coding model to determine the feature vectors corresponding to each historically answered question; wherein, the training method of the question coding model includes: obtaining several sets of second training datasets, each set of second training datasets including sample questions, and the average score rate and / or knowledge points corresponding to the sample questions; and training the second model to be trained based on several sets of second training datasets to obtain the question coding model.

[0016] Secondly, one embodiment of this application provides a knowledge point recommendation device, including: a first determining module, used to determine the weak knowledge points of a target user among multiple knowledge points; a second determining module, used to determine the learning vector of the target user for multiple knowledge points, wherein the learning vector represents the target user's mastery of multiple knowledge points; and a recommendation module, used to recommend superior knowledge points to the target user based on the target user's weak knowledge points among multiple knowledge points and the target user's learning vector for multiple knowledge points.

[0017] Thirdly, one embodiment of this application provides a computer-readable storage medium storing a computer program for performing the first aspect and the method described therein.

[0018] Fourthly, one embodiment of this application provides an electronic device, the electronic device comprising: a processor; a memory for storing processor-executable instructions; the processor being configured to perform the method described in the first aspect.

[0019] The knowledge point recommendation method provided in this application first identifies the target user's weak knowledge points among multiple knowledge points, then determines the target user's learning vector for these multiple knowledge points, and finally recommends superior knowledge points to the target user based on the target user's weak knowledge points and learning vector for these multiple knowledge points. Through the solution in this application, recommendations can be tailored to the target user's learning ability, making the knowledge point recommendation method more intelligent and personalized. This avoids problems such as recommended knowledge points not matching the target user's current learning level, leading to difficulties in solving knowledge points and long learning times. Correspondingly, the method in this application can improve the target user's learning benefits and enhance their learning confidence. Attached Figure Description

[0020] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 The diagram shown is a scenario applicable to an embodiment of this application.

[0022] Figure 2 The diagram shown is a flowchart illustrating a knowledge point recommendation method provided in an exemplary embodiment of this application.

[0023] Figure 3 The diagram shown is a flowchart illustrating the process of recommending superior learning knowledge points provided in an exemplary embodiment of this application.

[0024] Figure 4 The diagram shown is a schematic diagram of the recommended knowledge points provided by an exemplary embodiment of this application.

[0025] Figure 5 The diagram shown is a flowchart illustrating the process of determining weak knowledge points according to an exemplary embodiment of this application.

[0026] Figure 6The diagram shown is a flowchart illustrating the process of determining a learning vector according to an exemplary embodiment of this application.

[0027] Figure 7 The diagram shown is a schematic diagram of determining the learning vector of a target user according to an exemplary embodiment of this application.

[0028] Figure 8 The diagram shown is a schematic representation of the knowledge point recommendation device provided in an exemplary embodiment of this application.

[0029] Figure 9 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. Detailed Implementation

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Application Overview

[0032] In recent years, online learning and online answering have become increasingly common, especially during the last two years of online classes, with tens of millions of students attending classes, completing assignments, and taking exams online. Furthermore, more and more schools are establishing learning databases for each student. Meanwhile, some pioneering companies have long been collecting student learning data. With the accumulation of massive amounts of learning data, products combining cognitive intelligence core technologies with traditional education have been implemented. These products, based on traditional teaching methods, provide more targeted supplements to improve student performance. They can be said to improve academic performance while reducing student workload and pressure. For example, "personalized assignments" ensure that each student's assignments are different, directly addressing their weaknesses and reducing plagiarism, thus achieving the goal of reducing workload and increasing efficiency.

[0033] While many personalized recommendation products have emerged on the market, it remains to be seen whether these products truly achieve personalized diagnosis, personalized recommendations, and effective learning—that is, how to accurately answer the three questions of "what to learn, how to learn, and how to learn effectively" for students. For example, a "personalized learning manual" can only solve the problem of recommending similar questions to incorrect ones, but it cannot solve problems such as diagnosis and point-to-point recommendations. Moreover, it relies on original exam questions, so its personalization is entirely based on each individual's incorrect exam questions.

[0034] The relevant technical solutions applied in personalized education recommendation scenarios mainly fall into three categories: 1) "Knowledge Tracking," a small branch that has emerged in academic research in recent years. This type of solution mainly uses deep learning models to model students' abilities based on their historical chronological answer records, and then uses this model to predict the score rate of new questions. If recommending knowledge points, it is necessary to map questions to corresponding knowledge points for weighting. Since a question may have multiple knowledge points, there are some mapping errors in recommending knowledge points. 2) Similarly, based on students' historical learning, a student ability representation is modeled. Based on the student's ability representation, students with similar test scores and learning abilities are matched, and then the current / future weaknesses of similar students are recommended to the target user to improve learning outcomes. 3) Recommendation schemes are formed based on the teaching sequence formulated by teaching and research / subject matter experts and students' historical answer records. This approach is simple and has a low barrier to entry, and it is often seen in some traditional educational institutions. These institutions have strong teaching and research resources and design some convincing recommendation schemes based on experience. However, as traditional educational institutions gradually decrease and cannot meet the needs of personalized learning, this type of scheme is mostly used as a fallback solution for the two methods mentioned above, that is, an alternative when the model has no learning data.

[0035] While the aforementioned application schemes are usable in some scenarios, they have many obvious drawbacks. The first scheme currently forms an independent sub-field in scientific research and has some applications in some products, but its application in related products is limited or declining. This scheme has two main problems: first, it cannot solve the problem of sparse historical answers (commonly known as "cold start"), meaning that the prediction effect for points without answer records may have large deviations or central tendency effects; second, the scheme exhibits large fluctuations when new answer records are few, and becomes data insensitive when there is a lot of data. Although subsequent improvements have made similar time-weighted methods and given higher weight to nearby answers, they still fail to solve the problem of insensitivity to new answers and the inability to predict points with sparse answer records. The second approach involves identifying "similar users" to diagnose the target user's weaknesses and identify areas for priority recommendation and delayed recommendation. This approach effectively alleviates the situation where there are no or sparse answer records for recommended points. However, this approach is based on similar users, and data shows that similar users' learning progress will vary in the future. Furthermore, they are easily affected by thresholds during the recommendation process, resulting in recommended points / questions that users are weak in but cannot learn immediately, or in other words, they cannot learn them at all or their learning efficiency is low.

[0036] To address the limitations of the above solutions and draw upon experience in applying self-developed methods in the education field, this case proposes a knowledge point recommendation method. The core features of this method are as follows: First, it leverages a massive database of data from a personalized education mobile online reading data platform and precise teaching student answer data. This database covers years of student answer data from schools nationwide, including homework, weekly tests, monthly tests, midterms, finals, and joint exams. The quality, quantity, and richness of these data are fully guaranteed, providing a level of representation for subject-specific test questions that competitors cannot obtain, and is a prerequisite for the subsequent implementation of this experiment. Second, this case uses precise student profiling based on students' historical learning progress to identify users with similar scores and abilities. This similarity is reflected in overall school exam performance, similar mastery of various knowledge points, and similar learning abilities such as calculation ability, logical reasoning, and modeling ability. These abilities cannot be quantified, therefore, they can be abstractly represented by a 1024-dimensional high-dimensional dense vector. Thirdly, based on the second point, we can only find a group of users with similar scores and structures based on the current learning situation. However, these users are likely to diverge in the future. We select the characteristics of the progressive group and the regressive group according to the direction of the divergence. This solves the problem that the recommendation scheme is prone to falling into non-intelligent, non-personalized and recommending points / questions with large variance in answers. In other words, we avoid the drawbacks of the recommended knowledge points being difficult to solve and time-consuming, and realize the discovery of the weak points that are most likely to bring about score improvement within a limited learning time.

[0037] This case study addresses the two major issues of "diagnosis" and "recommendation." Based on historical data and millions of complete student learning records, it accurately models student ability representations. The modeling process incorporates the characteristics of future-improving groups among students with similar abilities, extracting highly differentiated and easily accessible advantageous knowledge points. This allows for the recommendation of knowledge points that most significantly improve grades. While helping students accurately pinpoint their weaknesses, it also effectively identifies areas for priority learning, delayed learning, and advanced expansion. This enhances learning efficiency, boosts learning confidence, and ultimately improves exam scores.

[0038] Exemplary scenario

[0039] The knowledge point recommendation method proposed in this application can be executed by an electronic device, which can be a terminal, such as a smartphone, tablet computer, desktop computer, etc.; or the electronic device can also be a server, such as an independent physical server, a server cluster composed of multiple servers, or a cloud server capable of cloud computing.

[0040] Based on the knowledge point recommendation method proposed in this application, this application embodiment provides a schematic diagram of the implementation environment of the knowledge point recommendation method. Figure 1 The diagram shown illustrates a scenario applicable to an embodiment of this application. Figure 1As shown, the application scenarios mentioned in the embodiments of this application include terminal 110 and server 120, and there is a communication connection between terminal 110 and server 120.

[0041] Terminal 110 can be a smartphone, tablet, desktop computer, etc. For example, terminal 110 has an application installed that can drive server 120 to execute the knowledge point recommendation method. Server 120 can be a physical machine or a virtual machine, and there can be one or more servers; this application embodiment does not limit the type or number of servers.

[0042] For example, terminal 110 determines the target user's weak knowledge points and learning vectors based on the target user's historical exam information, and sends the weak knowledge points and learning vectors to server 120. Server 120 determines the superior knowledge points based on the target user's learning vectors and weak knowledge points, and sends the corresponding encoding information of the superior knowledge points to terminal 110. Terminal 110 retrieves the practice questions corresponding to the knowledge points from the database based on the encoding information of the superior knowledge points, and presents the practice questions to the target user.

[0043] As an example of an online learning scenario, this approach can obtain structured information about student test questions, student behavior (e.g., time spent, waiting, hesitation, cheating), and scores in real time, adjusting student learning progress based on the online model. Furthermore, in online learning scenarios, resources are already within a fixed scope. By combining dynamic learning progress and the scope, it achieves real-time diagnosis of point mastery and completes real-time point ranking, recommending one or multiple points at a time, with multiple points dynamically changing after the previous point is completed.

[0044] As another example of an offline learning and review scenario, this scenario requires implementing multi-step recommendations at once. Specifically, it involves modeling a student's current profile based on available learning data before making recommendations, and then completing the recommendations based on a given review scope. This type of scenario recommends multiple points at once based on learning time, and then continues updating after the student completes their answers and the data is collected. This method is commonly used in school settings.

[0045] Exemplary methods

[0046] Figure 2 The diagram shown is a flowchart illustrating a knowledge point recommendation method provided in an exemplary embodiment of this application. Figure 2 As shown in the embodiments of this application, the knowledge point recommendation method includes the following steps.

[0047] Step S210: Identify the weak knowledge points of the target user among multiple knowledge points.

[0048] For example, the target user is a student, and a weak knowledge point refers to a test question for that knowledge point where the error rate exceeds a preset error threshold. Exams serve as a way to assess students' progress over time, and students are generally more attentive during exams, resulting in less data bias. Furthermore, exam questions are compiled by experienced teachers based on their regular teaching practices, ensuring high reliability and good discrimination. Therefore, this type of data is of paramount importance and includes data from common school weekly tests, monthly tests, midterms, finals, and joint exams. The original question stem is processed through word segmentation, characterization, and formulaization, and then, using a specific question coding model, a preliminary question representation (Ht) is obtained. This representation is used to identify multiple knowledge points corresponding to the questions, further pinpointing the target user's weak knowledge points within those areas.

[0049] Step S220: Determine the learning vectors for the target user for multiple knowledge points.

[0050] A learning vector represents a target user's mastery of multiple knowledge points. For example, a learning vector is a 1024-dimensional vector that uniquely represents the target user's abstract mastery of each knowledge point, as well as some abstractions from students' usual answers, knowledge point information, difficulty information, and learning behavior, such as students' problem-solving ability and students' serious attitude.

[0051] Step S230: Based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vectors for multiple knowledge points, recommend the best knowledge points to the target user.

[0052] While the above identifies weak knowledge points for target users, not all weak knowledge points need to be learned immediately. In other words, learning weak knowledge points requires a certain order. Blindly recommending them will place an excessive burden on target users and result in little learning benefit, which may lead to a loss of confidence and aversion to learning.

[0053] Specifically, based on the target user's learning vectors for multiple knowledge points, the system recommends the weaker knowledge points that will have the greatest impact on improving grades, and further recommends weaker knowledge points that are above the target user's current learning ability, progressing from easy to difficult in a step-by-step manner to avoid the target user falling into learning panic.

[0054] The knowledge point recommendation method provided in this application first identifies the target user's weak knowledge points among multiple knowledge points, then determines the target user's learning vectors for these multiple knowledge points, and finally recommends superior knowledge points to the target user based on these weak knowledge points and their learning vectors. This approach allows for tailored recommendations based on the target user's learning ability, making the knowledge point recommendation method more intelligent and personalized. It avoids problems such as recommended knowledge points not matching the target user's current learning level, leading to difficulties in solving knowledge points and long learning times. Consequently, the method in this application can improve the target user's learning benefits and enhance their learning confidence.

[0055] Figure 3 The diagram shown is a flowchart illustrating the recommended learning knowledge points provided in an exemplary embodiment of this application. Figure 2 Extending from the illustrated embodiment Figure 3 The illustrated embodiment will be described in detail below. Figure 3 The illustrated embodiments and Figure 2 The differences between the embodiments shown are not repeated here, and the similarities are not repeated here.

[0056] like Figure 3 As shown in the embodiments of this application, based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vectors for multiple knowledge points, the system recommends superior knowledge points to the target user, including the following steps.

[0057] Step S310: Determine the knowledge point discrimination level corresponding to the weak knowledge points.

[0058] Knowledge point differentiation represents the historical mastery information of weak knowledge points by users with similar scores and structures to the target user.

[0059] For example, based on the learning vectors of the target user for multiple knowledge points, the users with similar scores and structures corresponding to the target user are identified; based on the learning information of the users with similar scores and structures, the users who have improved and the users who have regressed are identified; based on the score information of the users who have improved on the weak knowledge points and the score information of the users who have regressed on the weak knowledge points, the knowledge point discrimination degree corresponding to the weak knowledge points is determined.

[0060] Specifically, users with similar learning vectors and learning structures refer to users whose learning vectors match those of the target user, meaning users with similar learning levels. For example, learning information refers to exam information for multiple knowledge points. By analyzing the exam information of users with similar learning vectors and learning structures for multiple knowledge points, we can identify users who have improved and those who have regressed. The difference between the score rates of improving and regressing users on their weak knowledge points is used to obtain the difference in mastery of each weak knowledge point, which is called the knowledge point discrimination index.

[0061] Step S320: Based on the learning vectors of the target user for multiple knowledge points, determine the knowledge point weakness degree corresponding to the weak knowledge points.

[0062] For example, feature vectors corresponding to multiple knowledge points are determined; based on the learning vectors of the target user for multiple knowledge points and the feature vectors corresponding to each knowledge point, a score prediction model is used to determine the knowledge point weakness corresponding to the weak knowledge point.

[0063] Knowledge point weakness represents the target user's future mastery of weak knowledge points, while the feature vector corresponding to each knowledge point represents the user's average mastery of that knowledge point. For example, the target user's learning vector and the feature vectors corresponding to multiple knowledge points are input into the score prediction model to predict the target user's future progress in multiple knowledge points, i.e., the knowledge point weakness. Specifically, the target user's learning vectors for multiple knowledge points and the corresponding feature vectors for each knowledge point are concatenated within the score prediction model.

[0064] Furthermore, the training method for the score prediction model includes: acquiring several sets of first training datasets, each set of first training datasets including sample questions and the masked question scores or unmasked question scores corresponding to the sample questions; and training the first model to be trained based on the several sets of first training datasets to obtain the score prediction model.

[0065] For example, a transfer learning approach is used to train the first model to be trained. Traditional training methods mask individual words in sample test questions, while this case assumes that the scores of each sample test question are independent of each other. The first model to be trained is trained by using the test scores corresponding to the masked sample test questions, thereby promoting a task-oriented training task.

[0066] For example, the loss function for the first model to be trained is:

[0067]

[0068] Where Zn represents the n sample questions in the sequence and the score corresponding to each sample question; t represents the position of the masked sample question. θ represents the predicted score of the masked sample question; M indicates that sample questions starting from position k are not visible to prevent premature information leakage; θ represents the training parameters of the first network model to be trained.

[0069] Figure 4 The diagram shown is an exemplary embodiment of the recommended learning knowledge points provided in this application. Figure 4As shown, in a certain first training dataset, the score of sample question Ht_3 is masked. Through the aforementioned loss function and the adjustment of the network parameters of the first training model during training, the first training network model can accurately output the score of Ht_3. The score output by the score prediction model can characterize the user's mastery of the knowledge points contained in the question; that is, the score prediction model can predict the user's weakness in the knowledge points contained in the question.

[0070] For example, the network structure of the score prediction model is a deep neural network (DNN). Furthermore, by inputting the target user's learning vector and the feature vector corresponding to the knowledge point into the DNN, the DNN can output the target user's knowledge weakness for that knowledge point. Further, by fusing the knowledge weakness and the knowledge point discrimination, the optimal knowledge points for the target user can be determined.

[0071] Further, based on several sets of first training datasets, the first model to be trained is trained to obtain the score prediction model, including: sorting several sets of first training datasets based on the answering time of sample questions in the first training datasets to obtain a training dataset sequence corresponding to several sets of first training datasets; and training the first model to be trained based on the training dataset sequence to obtain the score prediction model.

[0072] In this embodiment of the application, by sorting the answer times of sample test questions in the first training data and training the first training model based on the training dataset sequence, the score prediction model can better identify the progress of a particular student and their mastery of the knowledge points in the test questions.

[0073] Step S330: Based on the knowledge point differentiation and knowledge point weakness corresponding to the weak knowledge points, recommend superior learning knowledge points to the target user.

[0074] For example, based on the knowledge point differentiation and knowledge point weakness corresponding to weak knowledge points, recommending superior knowledge points to target users specifically includes: obtaining the difference data between the knowledge point differentiation and knowledge point weakness corresponding to weak knowledge points; determining the weight of the knowledge point differentiation corresponding to weak knowledge points based on the difference data; determining the weight of the knowledge point weakness corresponding to weak knowledge points based on the difference data; and recommending superior knowledge points to target users from among the weak knowledge points based on the weights of the knowledge point differentiation and knowledge point weakness.

[0075] For example, the knowledge point weakness is represented numerically; the higher the value, the greater the weakness. Similarly, the knowledge point differentiation is also represented numerically; the higher the value, the greater the differentiation. Furthermore, the smaller the difference between the knowledge point weakness and the knowledge point differentiation, the more likely the knowledge point is a weak point for the target user, but also that it is easy to learn. A preset equal difference threshold is used. If the difference between the knowledge point differentiation and the knowledge point weakness for a weak knowledge point is greater than the preset equal difference threshold, the weights of the two values ​​can be determined based on their specific values, with higher values ​​resulting in greater weights. If the difference between the knowledge point differentiation and the knowledge point weakness for a weak knowledge point is less than or equal to the preset equal difference threshold, it indicates that the weak knowledge point is easy for the target user to learn, but not a knowledge point that the target user urgently needs to fill; or the weak knowledge point urgently needs to be filled, but is not easy to learn. In this case, the weights corresponding to the knowledge point differentiation and the knowledge point weakness can be assigned according to preset recommendation rules. For example, if the preset recommendation rule is to first recommend easy-to-learn weak knowledge points to the target user, then the weight of knowledge point differentiation can be increased and the weight of knowledge point weakness can be decreased; or if the preset recommendation rule is to first recommend weak knowledge points that need to be filled to the user, then the weight of knowledge point differentiation can be decreased and the weight of knowledge point weakness can be increased.

[0076] For example, based on the weights of knowledge point differentiation and knowledge point weakness, the scores corresponding to weak knowledge points are obtained, and the weak knowledge points with the highest scores are identified as excellent knowledge points.

[0077] In this embodiment, the evaluation considers two dimensions: knowledge point differentiation and knowledge point weakness, to comprehensively assess whether a weak knowledge point should be recommended as a preferred knowledge point to the target user. That is, among numerous weak knowledge points, the target user's learning ability is used to prioritize the weak knowledge points that need to be learned, ensuring that priority knowledge points are addressed first, while more advanced and extended weak knowledge points are addressed later, in order to maximize the learning benefits for the target user.

[0078] Figure 5 The diagram shown is a schematic representation of a flowchart illustrating the process of determining weak knowledge points according to an exemplary embodiment of this application. Figure 2 Extending from the illustrated embodiment Figure 5 The illustrated embodiment will be described in detail below. Figure 5 The illustrated embodiments and Figure 2 The differences between the embodiments shown are not repeated here, and the similarities are not repeated here.

[0079] like Figure 5 As shown in the embodiments of this application, determining the target user's weak knowledge points among multiple knowledge points includes the following steps.

[0080] Step S510: Obtain the target user's exam score within a preset time period.

[0081] For example, the preset time period is one month. After obtaining the target user's learning vector, it is equivalent to understanding the target user's historical learning progress, learning ability, learning attitude, and other information. Furthermore, based on the given learning scope (such as the knowledge points learned this month), the target user's exam scores for the knowledge points learned this month are obtained.

[0082] Step S520: Based on the target user's exam score within a preset time period, determine the exam score of the reference user under the same exam conditions as the target user.

[0083] For example, "same exam conditions" refers to users who are in the same grade as the target user and use the same textbook version. That is, based on the target user's exam score, the current stage exam scores of users in the same grade and using the same textbook version are obtained.

[0084] Step S530: Based on the exam scores of the reference users and the exam scores of the target users, determine the users with the same scores as the target users from among the reference users.

[0085] For example, a tie-score threshold interval is constructed, and users with the same score as the target user are identified based on this interval. Assume the tie-score threshold interval is [-10%, 10%], meaning users whose exam scores are within ±10% of the target user's score are considered to be users with the same score as the target user.

[0086] Step S540: Based on the historical exam information of users with the same score, identify the weak knowledge points of the target user in multiple knowledge areas.

[0087] For example, historical exam information includes midterm exams, final exams, monthly exams, and weekly exams. Based on the historical exam information of users with the same score over a period of time, the weak knowledge points of these users are identified, and these weak knowledge points are then used as the weak knowledge points of the target user.

[0088] As a preferred example, based on the historical exam information of users with the same score, the weak knowledge points of the target user in multiple knowledge points are determined, including: determining the learning vectors of users with the same score for multiple knowledge points based on the historical exam information of users with the same score; determining the users with the same score and similarity to the learning vectors of users with the same score for multiple knowledge points and the learning vectors of the target user for multiple knowledge points based on the similarity between the learning vectors of users with the same score and similarity to the learning vectors of the target user for multiple knowledge points; and determining the weak knowledge points of the target user in multiple knowledge points based on the historical exam information of users with the same score and similarity to the learning vectors of the target user.

[0089] Specifically, based on the historical exam information of users with the same score, learning vectors for multiple knowledge points are determined for each user, thus defining their learning ability representation. Further, the similarity between the learning vectors of the users with the same score and the learning vector of the target user is calculated; for example, cosine similarity is used. Users with the same score are ranked according to their similarity scores, and the top K users in the ranking are selected as the target user's corresponding isomorphic users. Further, based on the historical exam information of these isomorphic users, their corresponding weak knowledge points are identified, and these weak knowledge points are then used as the target user's weak knowledge points.

[0090] In an exemplary embodiment of this application, the determination of the target user's weak knowledge points among multiple knowledge points is based on the historical exam information corresponding to users with similar scores and structures. This includes: determining the user's mastery of multiple knowledge points based on the historical exam information of users with similar scores and structures; and for each knowledge point among the multiple knowledge points, if a preset number of users with similar scores and structures have a mastery of the knowledge point that meets the weak knowledge point setting conditions, then the knowledge point is determined as the target user's weak knowledge point.

[0091] For example, the score rate of users with similar scores and structures on multiple knowledge points can be determined by analyzing their historical exam information. This score rate represents the degree of mastery of these users over these knowledge points. Assuming the preset number is 70% of the total number of users with similar scores and structures, and the condition for identifying weak knowledge points is a score rate of less than 60% for that knowledge point. That is, for each knowledge point, if 70% of users with similar scores and structures have a score rate of less than 60% for that knowledge point, then that knowledge point is identified as a weak knowledge point for those users, and further identified as a weak knowledge point for the target user.

[0092] In this embodiment of the application, the weak knowledge points of the target user are determined by using the historical exam information of users with the same score as the target user, preferably by using the historical exam information of users with the same score and similar structure as the target user. The solution in this embodiment of the application can more fully and comprehensively determine the weak knowledge points of the target user.

[0093] Figure 6 The diagram shown is a schematic representation of the process for determining a learning vector according to an exemplary embodiment of this application. Figure 2 Extending from the illustrated embodiment Figure 6 The illustrated embodiment will be described in detail below. Figure 6 The illustrated embodiments and Figure 2 The differences between the embodiments shown are not repeated here, and the similarities are not repeated here.

[0094] like Figure 6 As shown in the embodiments of this application, determining the learning vector for a target user for multiple knowledge points includes the following steps.

[0095] Step S610: Obtain the target user's historical test questions.

[0096] Step S620: Determine the feature vectors corresponding to each of the historical test questions.

[0097] The feature vectors of historical test questions represent the target user's answering time, score information and average score rate, difficulty information of historical test questions, and the knowledge points corresponding to historical test questions.

[0098] For example, determining the feature vectors corresponding to each historically answered test question includes: using a test question coding model to determine the feature vectors corresponding to each historically answered test question; wherein, the training method of the test question coding model includes: obtaining several sets of second training datasets, each set of second training datasets including sample test questions, and the average score rate and / or knowledge points corresponding to the sample test questions; and training the second model to be trained based on several sets of second training datasets to obtain the test question coding model.

[0099] Specifically, the test question coding model is used to process historical test questions through word segmentation, characterization, and formulaization to obtain the feature vectors corresponding to each historical test question.

[0100] Here, based on common downstream scenarios and to better serve downstream businesses, this embodiment of the application designs several sub-tasks related to downstream task scenarios and incorporates them into the training of the question coding model, forming a multi-task learning approach for answering question representation learning. For example, for each set of sample questions in the second training dataset, a question difficulty prediction task is trained based on the obtained average score rate for each question. Multi-label or multi-class prediction tasks can also be constructed, such as predicting the knowledge points or anchor points of the questions. During the multi-task learning process, more downstream scenario information can be incorporated, resulting in better generalization of the question's feature vectors.

[0101] Step S630: Based on the feature vectors corresponding to each of the historical test questions, determine the learning vectors for the target user for multiple knowledge points.

[0102] Figure 7 The diagram illustrates a method for determining a target user's learning vector according to an exemplary embodiment of this application. Specifically, see [link to relevant documentation]. Figure 7Ht_1-Ht_n represent the feature vectors of historical test questions; for example, Ht_1 corresponds to features 1_1, 1_2, 1_3, 1_4, etc. Further, the knowledge points, time consumption, difficulty, and score rate of each historical test question are aligned with the features of the historical test questions, and these features are input into the deep knowledge tracking model. This model further outputs the learning vectors of multiple knowledge points included in the target user's historical test questions, as well as more detailed information on the mastery level of each knowledge point included in the historical test questions.

[0103] Specifically, the historical answers of students are serialized according to their occurrence time and aligned with the corresponding question features. Alternatively, a question resource dictionary could be constructed, representing questions by their index numbers, but this is only suitable for closed question sets. In contrast, using feature vectors to represent historical answers is more flexible and has a wider range of applications.

[0104] Furthermore, the target user's test scores, average score rate, difficulty, behavioral characteristics, knowledge points or anchor points, and historical test answers are vectorized and aligned with the corresponding answer sequences of historical test answers. This vectorized data is then input into a deep knowledge tracing model, such as an end-to-end Transformer, to predict the target user's mastery level at each point. The vector of the layer preceding the prediction output layer of the deep knowledge tracing model—that is, the last hidden layer vector of the deep knowledge tracing model—is extracted and used as the target user's learning vector.

[0105] The above text combined Figures 2 to 7 The method embodiments of this application are described in detail below, in conjunction with... Figure 8 The present application provides a detailed description of the apparatus embodiments. It should be understood that the descriptions of the method embodiments correspond to the descriptions of the apparatus embodiments; therefore, any parts not described in detail can be found in the foregoing method embodiments.

[0106] Figure 8 The diagram shown is a structural schematic of a knowledge point recommendation device provided in an exemplary embodiment of this application. Figure 8 As shown, the knowledge point recommendation device 80 provided in this embodiment includes:

[0107] The first determination module 810 is used to determine the target user's weak knowledge points among multiple knowledge points;

[0108] The second determining module 820 is used to determine the learning vector of the target user for multiple knowledge points. The learning vector represents the target user's mastery of multiple knowledge points.

[0109] The recommendation module 830 is used to recommend superior knowledge points to the target user based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vector for multiple knowledge points.

[0110] In one embodiment of this application, the recommendation module 830 is further configured to: determine the knowledge point discrimination degree corresponding to the weak knowledge point, wherein the knowledge point discrimination degree represents the historical mastery information of the weak knowledge point by users with similar scores and structures to the target user; determine the knowledge point weakness degree corresponding to the weak knowledge point based on the learning vector of the target user for multiple knowledge points, wherein the knowledge point weakness degree represents the target user's future mastery information of the weak knowledge point; and recommend superior knowledge points to the target user based on the knowledge point discrimination degree and the knowledge point weakness degree corresponding to the weak knowledge point.

[0111] In one embodiment of this application, the recommendation module 830 is further configured to: obtain the difference data between the knowledge point discrimination degree and the knowledge point weakness degree corresponding to the weak knowledge point; determine the weight of the knowledge point discrimination degree corresponding to the weak knowledge point based on the difference data; determine the weight of the knowledge point weakness degree corresponding to the weak knowledge point based on the difference data; and recommend superior knowledge points to the target user from the weak knowledge points based on the weight of the knowledge point discrimination degree and the weight of the knowledge point weakness degree.

[0112] In one embodiment of this application, the recommendation module 830 is further configured to: determine the users with similar scores and structures to the target user based on the target user's learning vectors for multiple knowledge points; determine the improved users and the regressing users among the users with similar scores and structures based on the learning information corresponding to the users with similar scores and structures; and determine the knowledge point discrimination degree corresponding to the weak knowledge points based on the score information of the improved users on the weak knowledge points and the score information of the regressing users on the weak knowledge points.

[0113] In one embodiment of this application, the recommendation module 830 is further configured to determine the feature vectors corresponding to each of the multiple knowledge points, wherein the feature vectors corresponding to the knowledge points represent the average mastery of the knowledge points by the user; and based on the learning vectors of the target user for the multiple knowledge points and the feature vectors corresponding to the multiple knowledge points, use a score prediction model to determine the weakness of the knowledge points corresponding to the weak knowledge points.

[0114] In one embodiment of this application, the recommendation module 830 is further configured to train the score prediction model by: acquiring several sets of first training datasets, each set of first training datasets including sample questions and the masked question score or the unmasked question score corresponding to the sample questions; and training the first model to be trained based on the several sets of first training datasets to obtain the score prediction model.

[0115] In one embodiment of this application, the recommendation module 830 is further configured to sort several sets of the first training dataset based on the answering time of sample questions in the first training dataset to obtain a training dataset sequence corresponding to several sets of the first training dataset; and train the first model to be trained based on the training dataset sequence to obtain a score prediction model.

[0116] In one embodiment of this application, the first determining module 810 is further configured to: obtain the exam score of the target user within a preset time period; determine the exam score of a reference user under the same exam conditions based on the exam score of the target user within the preset time period; determine the user with the same score as the target user from among the reference users based on the exam scores of the reference users and the exam score of the target user; and determine the weak knowledge points of the target user among multiple knowledge points based on the historical exam information corresponding to the user with the same score.

[0117] In one embodiment of this application, the first determining module 810 is further configured to: determine the learning vectors of the users with the same score for multiple knowledge points based on their historical exam information; determine the users with the same score and similarity to the target user based on the similarity between the learning vectors of the users with the same score for multiple knowledge points and the learning vectors of the target user for multiple knowledge points; and determine the weak knowledge points of the target user among multiple knowledge points based on the historical exam information of the users with the same score and similarity.

[0118] In one embodiment of this application, the first determining module 810 is further configured to determine the mastery of multiple knowledge points by users with similar scores and structures based on their historical exam information; and for each of the multiple knowledge points, if a preset number of users with similar scores and structures have a mastery of the knowledge point that meets the conditions for setting weak knowledge points, then the knowledge point is determined as the weak knowledge point of the target user.

[0119] In one embodiment of this application, the second determining module 820 is further configured to: obtain the target user's historical test questions; determine the feature vectors corresponding to each of the historical test questions, wherein the feature vectors represent the target user's answering time, score information and average score rate, difficulty information of the historical test questions and knowledge points corresponding to the historical test questions; and determine the target user's learning vectors for multiple knowledge points based on the feature vectors corresponding to each of the historical test questions.

[0120] In one embodiment of this application, the second determining module 820 is further configured to determine the feature vectors corresponding to each historically answered question using a question coding model; wherein, the training method of the question coding model includes: acquiring several sets of second training datasets, each set of second training datasets including sample questions, and the average score rate and / or knowledge points corresponding to the sample questions; and training the second model to be trained based on the several sets of second training datasets to obtain the question coding model.

[0121] Below, for reference Figure 9 This describes an electronic device according to embodiments of the present application. Figure 9 The diagram shown is a structural schematic of an electronic device provided in an exemplary embodiment of this application.

[0122] like Figure 9As shown, the electronic device 90 includes one or more processors 901 and memory 902.

[0123] The processor 901 may be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and may control other components in the electronic device 90 to perform desired functions.

[0124] The memory 902 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 901 may execute the program instructions to implement the methods of the various embodiments of this application described above and / or other desired functions. The computer-readable storage medium may also store various content such as weak knowledge points, learning vectors of target users, excellent knowledge points, multiple knowledge points, knowledge point discrimination, and knowledge point weakness.

[0125] In one example, the electronic device 90 may also include an input device 903 and an output device 904, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0126] The input device 903 may include, for example, a keyboard, a mouse, etc.

[0127] The output device 904 can output various information to the outside, including weak knowledge points, the target user's learning vector, excellent knowledge points, multiple knowledge points, knowledge point discrimination, and knowledge point weakness. The output device 904 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0128] Of course, for the sake of simplicity, Figure 9 Only some of the components of the electronic device 90 relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device 90 may include any other suitable components depending on the specific application.

[0129] In addition to the methods and apparatus described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps of the methods described above according to various embodiments of this application.

[0130] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0131] Furthermore, embodiments of this application may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform the steps of the methods described above according to various embodiments of this application.

[0132] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0133] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0134] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0135] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0136] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0137] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A knowledge point recommendation method, characterized in that, include: Identify the target users' weak knowledge points across multiple knowledge areas; Determine the learning vector for the target user for the multiple knowledge points, wherein the learning vector represents the target user's mastery of the multiple knowledge points; Based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vector for the multiple knowledge points, recommend the best knowledge points to the target user. The optimal learning knowledge points are determined based on the knowledge point discrimination degree and the knowledge point weakness degree. The knowledge point discrimination degree represents the historical mastery information of the weak knowledge points by users with similar scores and structures to the target user. The users with similar scores and structures are users whose learning vectors match those of the target user. The knowledge point weakness level represents the target user's future mastery of the weak knowledge points.

2. The method according to claim 1, characterized in that, The step of recommending superior learning knowledge points to the target user based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vector for the multiple knowledge points includes: Determine the knowledge point discrimination index corresponding to the weak knowledge points; Based on the learning vectors of the target user for the multiple knowledge points, the knowledge point weakness degree corresponding to the weak knowledge point is determined. Based on the knowledge point differentiation and knowledge point weakness corresponding to the weak knowledge points, the superior learning knowledge points are recommended to the target user.

3. The method according to claim 2, characterized in that, The process of recommending superior learning knowledge points to the target user based on the knowledge point differentiation and knowledge point weakness corresponding to the weak knowledge points includes: Obtain the difference data between the knowledge point discrimination degree and the knowledge point weakness degree corresponding to the weak knowledge points; Based on the difference data, determine the weight of the knowledge point discrimination degree corresponding to the weak knowledge point; Based on the difference data, the weight of the weakness of the knowledge point corresponding to the weak knowledge point is determined; Based on the weights of the knowledge point differentiation and the knowledge point weakness, the superior knowledge points are recommended to the target user from the weak knowledge points.

4. The method according to claim 2, characterized in that, Determining the knowledge point discrimination index corresponding to the weak knowledge point includes: Based on the learning vectors of the target user for the multiple knowledge points, determine the isomorphic users corresponding to the target user; Based on the learning information corresponding to the homogeneous users, the progressive users and regressive users among the homogeneous users are determined; Based on the scores of the improved users on the weak knowledge points and the scores of the regressed users on the weak knowledge points, the knowledge point discrimination degree corresponding to the weak knowledge points is determined.

5. The method according to claim 2, characterized in that, The step of determining the knowledge point weakness degree corresponding to the weak knowledge point based on the learning vector of the target user for the multiple knowledge points includes: Determine the feature vector corresponding to each of the multiple knowledge points, and the feature vector corresponding to each knowledge point represents the user's average mastery of the knowledge point; Based on the target user's learning vectors for the multiple knowledge points and the feature vectors corresponding to each of the multiple knowledge points, a score prediction model is used to determine the knowledge point weakness corresponding to the weak knowledge point.

6. The method according to claim 5, characterized in that, The training method for the score prediction model includes: Obtain several sets of first training datasets. Each set of first training datasets includes sample questions and the masked question score or the unmasked question score corresponding to the sample questions. Based on the aforementioned sets of first training datasets, the first model to be trained is trained to obtain the score prediction model.

7. The method according to claim 6, characterized in that, The step of training the first model to be trained based on the plurality of first training datasets to obtain the score prediction model includes: Based on the answering time of sample questions in the first training dataset, the plurality of first training datasets are sorted to obtain the training dataset sequence corresponding to the plurality of first training datasets; Based on the training dataset sequence, the first model to be trained is trained to obtain the score prediction model.

8. The method according to claim 1, characterized in that, The process of identifying the target user's weak knowledge points across multiple knowledge areas includes: Obtain the target user's exam scores within a preset time period; Based on the target user's exam score within a preset time period, determine the exam score of a reference user under the same exam conditions as the target user. Based on the exam scores of the reference users and the exam scores of the target user, identify the users with the same score as the target user from among the reference users; Based on the historical exam information of the users with the same score, the weak knowledge points of the target user in multiple knowledge areas are identified.

9. The method according to claim 8, characterized in that, The step of determining the target user's weak knowledge points across multiple knowledge areas based on the historical exam information of the users with the same score includes: Based on the historical exam information of the users with the same score, the learning vectors for the multiple knowledge points of the users with the same score are determined; Based on the similarity between the learning vectors of the same-score users for the multiple knowledge points and the learning vectors of the target user for the multiple knowledge points, the same-score and isomorphic users corresponding to the target user are determined from the same-score users; Based on the historical exam information of the users with similar scores and structures, the weak knowledge points of the target user in multiple knowledge areas are identified.

10. The method according to claim 9, characterized in that, The step of determining the target user's weak knowledge points across multiple knowledge areas based on the historical exam information of the users with similar scores and structures includes: Based on the historical exam information of the users with similar scores and structures, determine the degree of mastery of the users with similar scores and structures over the multiple knowledge points; For each of the multiple knowledge points, if a preset number of users with similar knowledge levels meet the criteria for setting weak knowledge points, then the knowledge point is identified as a weak knowledge point for the target user.

11. The method according to claim 1, characterized in that, Determining the learning vector for the target user regarding the multiple knowledge points includes: Obtain the target user's historical test questions; Determine the feature vector corresponding to each of the historical test questions, wherein the feature vector represents the target user's answering time, score information and average score rate, difficulty information of the historical test questions, and knowledge points corresponding to the historical test questions; Based on the feature vectors corresponding to each of the historical test questions, the learning vectors for the target user for the multiple knowledge points are determined.

12. The method according to claim 11, characterized in that, Determining the feature vectors corresponding to each of the historically answered questions includes: Using a question coding model, the feature vectors corresponding to each of the historically answered questions are determined; wherein, the training method of the question coding model includes: Obtain several sets of second training datasets, each set of second training datasets including sample test questions, and the average score rate and / or knowledge points corresponding to the sample test questions; Based on the aforementioned sets of second training datasets, the second model to be trained is trained to obtain the test item coding model.

13. A knowledge point recommendation device, characterized in that, include: The first identification module is used to identify the target user's weak knowledge points among multiple knowledge points; The second determining module is used to determine the learning vector of the target user for the multiple knowledge points, wherein the learning vector represents the target user's mastery of the multiple knowledge points; The recommendation module is used to recommend superior knowledge points to the target user based on the target user's weak knowledge points in multiple knowledge points and the target user's learning vector for the multiple knowledge points; The optimal learning knowledge points are determined based on the knowledge point discrimination degree and the knowledge point weakness degree. The knowledge point discrimination degree represents the historical mastery information of the weak knowledge points by users with similar scores and structures to the target user. The users with similar scores and structures are users whose learning vectors match those of the target user. The knowledge point weakness level represents the target user's future mastery of the weak knowledge points.

14. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the knowledge point recommendation method according to any one of claims 1 to 12.

15. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is used to execute the knowledge point recommendation method according to any one of claims 1 to 12.

Images