User exercise recommendation information generation method and device, electronic equipment and readable medium

Abstract

The embodiment of the invention discloses a user exercise recommendation information generation method and device, electronic equipment and a readable medium. A specific embodiment of the method comprises the steps of obtaining a user exercise interaction data sequence and an exercise concept association data sequence; based on each exercise answering data sequence and each exercise concept associated data sequence, generating user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed conversion graph data; inputting the user exercise interaction data sequence, the user exercise heterogeneous graph data, the exercise concept hypergraph data and the exercise directed conversion graph data into a knowledge tracking model to obtain each user knowledge state vector; for each learning user, performing detection processing on the user knowledge state vector to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; generating a second knowledge state post-concept information set; and generating user exercise recommendation information. According to the embodiment, the user exercise recommendation information more adaptive to the user demand can be generated.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method, device, electronic device, and readable medium for generating user exercise recommendation information. Background Art

[0002] With the explosive growth of online learning resources (such as course explanation resources and exercise resources), learners usually find it difficult to choose learning resources suitable for themselves from the massive amount of learning resources for learning. Therefore, how to recommend personalized learning resources to learning users to guide their learning has become an important exploration direction for online learning platforms. At present, when recommending exercise resources to learning users, the usual method is: according to the learning user's historical answering behavior, through the knowledge tracking technology based on the relationship between exercises and conceptual knowledge points, the learning user's knowledge status at the current learning stage is determined, and the exercises related to the unmastered knowledge points involved in the current knowledge status are used as user exercise recommendation information, and pushed to the learning user for learning.

[0003] However, in practice, it is found that when the above method is used to recommend exercise resources to learning users, the following technical problems often occur: First, because the knowledge tracking technology based on the relationship between exercises and concept knowledge points is centered on exercises and concept knowledge points, it ignores the learning users' preferences for exercises, making the generated user exercise recommendation information less compatible with user needs, thus making it difficult to push user exercise recommendation information that is more compatible with user needs to learning users; Second, since the exercises recommended based on the knowledge tracking results can only cover the knowledge points learned in the historical period, the impact of the learner's learning path on the user's learning needs is ignored, making the generated user exercise recommendation information less adaptable to user needs, which makes it difficult to push user exercise recommendation information that is more adaptable to user needs to learning users.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure concept and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the invention

[0005] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0006] Some embodiments of the present disclosure propose a method, device, electronic device, and readable medium for generating user exercise recommendation information to solve one or more of the technical problems mentioned in the above background technology section.

[0007] In a first aspect, some embodiments of the present disclosure provide a method for generating user exercise recommendation information, the method comprising: obtaining a user exercise interaction data sequence and an exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the above user exercise interaction data sequence comprises an exercise answer data sequence; based on each exercise answer data sequence and the above exercise concept association data sequence included in the above user exercise interaction data sequence, generating user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data; inputting the above user exercise interaction data sequence, the above user exercise heterogeneous graph data, the above exercise concept hypergraph data and the above exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in the above each user knowledge state vector is consistent with the above each learning user. Corresponding to the learning users in the users, the above-mentioned knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; for each of the above-mentioned learning users, the following steps are performed: each component in the user knowledge state vector corresponding to the above-mentioned learning user is detected and processed to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; based on the above-mentioned second knowledge state component information sequence and the preset concept learning path information set, a second knowledge state post-concept information set is generated; based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, user exercise recommendation information is generated, and the above-mentioned user exercise recommendation information is stored in the cloud server for the above-mentioned learning user to obtain.

[0008] In a second aspect, some embodiments of the present disclosure provide a device for generating user exercise recommendation information, the device comprising: an acquisition unit, configured to acquire a user exercise interaction data sequence and an exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the above user exercise interaction data sequence comprises an exercise answer data sequence; a generation unit, configured to generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data based on each exercise answer data sequence and the above exercise concept association data sequence included in the above user exercise interaction data sequence; an input unit, configured to input the above user exercise interaction data sequence, the above user exercise heterogeneous graph data, the above exercise concept hypergraph data and the above exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in the above each user knowledge state vector The quantity corresponds to the learning users among the above-mentioned learning users, and the above-mentioned knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; the execution unit is configured to perform the following steps for each learning user among the above-mentioned learning users: detect and process each component in the user knowledge state vector corresponding to the above-mentioned learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; based on the above-mentioned second knowledge state component information sequence and the preset concept learning path information set, generate a second knowledge state post-concept information set; based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, generate user exercise recommendation information, and store the above-mentioned user exercise recommendation information to the cloud server for the above-mentioned learning user to obtain.

[0009] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner of the above-mentioned first aspect.

[0010] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any implementation manner of the above-mentioned first aspect is implemented.

[0011] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the user exercise recommendation information generation method of some embodiments of the present disclosure, the user exercise recommendation information with a higher degree of adaptability to user needs can be pushed to the learning user. Specifically, the reason for pushing the user exercise recommendation information with a higher degree of adaptability to user needs to the learning user is that: due to the knowledge tracking technology based on the relationship between exercises and concept knowledge points, the center is exercises and concept knowledge points, and the learning user's preference for exercises is ignored, so that the generated user exercise recommendation information has a low degree of adaptability to user needs, thereby making it difficult to push the user exercise recommendation information with a higher degree of adaptability to user needs to the learning user. Based on this, the user exercise recommendation information generation method of some embodiments of the present disclosure, first, obtains the user exercise interaction data sequence and the exercise concept association data sequence corresponding to each learning user. Among them, each user exercise interaction data in the above-mentioned user exercise interaction data sequence includes an exercise answer data sequence. Thus, the historical exercise answering behavior of each learning user and the association relationship between exercises and concept knowledge points can be obtained. Secondly, based on the various exercise answer data sequences and the exercise concept association data sequences included in the above user exercise interaction data sequence, user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data are generated. Thus, high-order relationships between learning users and exercises, between exercises and concept knowledge points, and between exercises can be obtained. Then, the above user exercise interaction data sequence, the above user exercise heterogeneous graph data, the above exercise concept hypergraph data and the above exercise directed transition graph data are input into the pre-trained knowledge tracking model to obtain each user knowledge state vector. Among them, the user knowledge state vector in the above each user knowledge state vector corresponds to the learning user in the above each learning user, and the above knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module. Thus, through the knowledge tracking model, according to the connection between the learning user, the exercise and the concept, a knowledge state vector representing the user's knowledge mastery degree can be constructed for each learning user. Finally, for each of the above-mentioned learning users, the following steps are performed: each component in the user knowledge state vector corresponding to the above-mentioned learning user is detected and processed to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; based on the above-mentioned second knowledge state component information sequence and the preset concept learning path information set, a second knowledge state post-concept information set is generated; based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, user exercise recommendation information is generated, and the above-mentioned user exercise recommendation information is stored in the cloud server for the above-mentioned learning user to obtain. In this way, for each learning user, exercises related to the concept knowledge points that have not been mastered can be pushed to the learning user.Therefore, the method for generating user exercise recommendation information in some embodiments of the present disclosure can introduce students' preference information for exercises into the knowledge tracking model by inputting the user exercise heterogeneous graph data representing the high-order relationship between students and exercises into the knowledge tracking model. Furthermore, by integrating the relationship between students, exercises and concepts, a more accurate knowledge state can be generated for learning users. Thus, user exercise recommendation information that is more suitable for user needs can be generated. Furthermore, the number of recommendations for exercises with low adaptability can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0013] Figure 1 is a flow chart of some embodiments of the method for generating user exercise recommendation information according to the present disclosure; Figure 2 is a schematic diagram of the structure of some embodiments of the device for generating user exercise recommendation information according to the present disclosure; Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0014] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0015] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0016] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0017] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0018] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0019] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0020] Figure 1 The process 100 of some embodiments of the method for generating user exercise recommendation information according to the present disclosure is shown. The method for generating user exercise recommendation information comprises the following steps: Step 101: Obtain user exercise interaction data sequences and exercise concept association data sequences corresponding to each learning user.

[0021] In some embodiments, the execution subject (e.g., computing device) of the method for generating user exercise recommendation information may obtain user exercise interaction data sequences and exercise concept association data sequences corresponding to each learning user from the database through a wired connection or a wireless connection. Among them, each user exercise interaction data in the above user exercise interaction data sequence may include but is not limited to an exercise answer data sequence. Each user exercise interaction data in the above user exercise interaction data sequence corresponds one-to-one to a learning user among the above learning users. The learning user may be a user who logs in to the course learning platform to practice exercises. Each learning user has an associated learning user identifier. The above learning user identifier may be a unique identifier of the learning user. The above exercise answer data sequence may be a sequence of exercise answer data arranged from early to late in the order of answer time. The exercise answer data in the above exercise answer data sequence may include an exercise identifier and an answer category identifier. Among them, the above exercise identifier may be a unique identifier of the exercise. The above answer category identifier may be an identifier that characterizes whether the answer to the corresponding exercise is correct. For example, the above-mentioned answer category identifier can be 1 or 0. When the above-mentioned answer category identifier is 1, it indicates that the answer to the corresponding exercise is correct. When the above-mentioned answer category identifier is 0, it indicates that the answer to the corresponding exercise is wrong. The exercise concept association data in the above-mentioned exercise concept association data sequence can represent the relationship between the examined concept knowledge point and the exercise. The exercise concept association data in the above-mentioned exercise concept association data sequence may include a concept identifier and an exercise identifier group. The above-mentioned concept identifier can be a unique identifier of a concept knowledge point. It should be pointed out that the above-mentioned wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0022] Step 102, based on each exercise answer data sequence and exercise concept association data sequence included in the user exercise interaction data sequence, generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data.

[0023] In some embodiments, the execution subject may generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data in various ways based on each exercise answer data sequence and the exercise concept association data sequence included in the user exercise interaction data sequence. Among them, the user exercise heterogeneous graph data may be adjacency matrix data representing the user exercise heterogeneous graph. The user exercise heterogeneous graph may include positive heterogeneous graph and negative heterogeneous graph. The positive heterogeneous graph may be a bipartite graph composed of each learning user and each exercise to which the learning user has a correct answer. The negative heterogeneous graph may be a bipartite graph composed of each learning user and each exercise to which the learning user has a wrong answer. The exercise concept hypergraph data may be an association matrix representing the exercise concept hypergraph. The exercise concept hypergraph may be a hypergraph with exercises as nodes and concepts associated with exercises as hyperedges. The exercise directed transition graph data may be adjacency matrix data representing the exercise directed transition graph. The exercise directed conversion graph may be a directed graph with exercises as nodes and the order of exercises as directed edges, wherein the directed edges may be edges from the node corresponding to the exercise answered first by the user to the node corresponding to the exercise answered later by the user.

[0024] In some optional implementations of some embodiments, the execution subject may generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data based on each exercise answer data sequence included in the user exercise interaction data sequence and the exercise concept association data sequence through the following steps: The first step is to classify and process the various exercise answer data included in the above-mentioned user exercise interaction data sequence to obtain the correct answer exercise data sequence and the incorrect answer exercise data sequence corresponding to each learning user. Among them, the correct answer exercise data in the above-mentioned correct answer exercise data sequence corresponds one-to-one to the learning users among the above-mentioned learning users. The correct answer exercise data in the above-mentioned correct answer exercise data sequence can be a sequence of data of each exercise correctly answered by the corresponding learning user. The incorrect answer exercise data in the above-mentioned incorrect answer exercise data sequence corresponds one-to-one to the learning users among the above-mentioned learning users. The incorrect answer exercise data in the above-mentioned incorrect answer exercise data sequence can be a sequence of data of each exercise incorrectly answered by the corresponding learning user. For each exercise answer data sequence included in the above-mentioned user exercise interaction data sequence, perform the following sub-steps: In the first sub-step, the exercise answer data satisfying the preset correct answer condition are selected in sequence from the exercise answer data sequence as the first interactive exercise data, and the first interactive exercise data sequence is obtained as the correct answer exercise data. The preset correct answer condition may be that the answer category identifier included in the exercise answer data is the same as the preset correct category identifier. The preset correct category identifier may be a preset identifier indicating that the exercise answer is correct.

[0025] The second sub-step is to delete each exercise answer data in the exercise answer data sequence that matches the first interactive exercise data sequence, and obtain a second interactive exercise data sequence as incorrect answer exercise data. Wherein, matching the first interactive exercise data sequence can be: the exercise answer data in the exercise answer data sequence is the same as any first interactive exercise data in the first interactive exercise data sequence.

[0026] The second step is to generate user exercise heterogeneous graph data corresponding to each learning user based on the above correct answer exercise data sequence and the above incorrect answer exercise data sequence. The above execution subject can generate user exercise heterogeneous graph data corresponding to each learning user based on the above correct answer exercise data sequence and the above incorrect answer exercise data sequence in various ways.

[0027] In some optional implementations of some embodiments, the execution subject may generate user exercise heterogeneous graph data corresponding to each learning user based on the correct answer exercise data sequence and the incorrect answer exercise data sequence through the following steps: Step one, determine the first column identifier sequence corresponding to the above-mentioned correctly answered exercise data sequence. Among them, the first column identifier in the above-mentioned first column identifier sequence can be the identifier of the correctly answered exercise. First, each exercise identifier corresponding to each correctly answered exercise data in the above-mentioned correctly answered exercise data sequence is determined as a correctly answered exercise identifier group to obtain a correctly answered exercise identifier group sequence. Then, each correctly answered exercise identifier in the above-mentioned correctly answered exercise identifier group sequence is de-redundanted to obtain a correctly answered exercise identifier set after de-redundancy as the first column identifier set. Finally, the first column identifier in the above-mentioned first column identifier set is sorted by a preset sorting algorithm to obtain a first column identifier sequence. For example, the above-mentioned sorting algorithm can be but is not limited to one of the following: bubble sort, quick sort.

[0028] Step 2, determine the second column identification sequence corresponding to the above incorrectly answered exercise data sequence. Among them, the second column identification in the above second column identification sequence can be the identification of the incorrectly answered exercise. First, each exercise identification corresponding to each incorrectly answered exercise data in the above incorrectly answered exercise data sequence is determined as an incorrectly answered exercise identification group to obtain an incorrectly answered exercise identification group sequence. Then, each incorrectly answered exercise identification in the above incorrectly answered exercise identification group sequence is de-redundanted to obtain the incorrectly answered exercise identification set after de-redundancy as the second column identification set. Finally, the second column identification in the above second column identification set is sorted by the above sorting algorithm to obtain a second column identification sequence.

[0029] Step 3: Determine the row identification sequence corresponding to each of the above-mentioned learning users. The row identification in the above-mentioned row identification sequence may be the identification of the learning user. First, the learning user identification corresponding to each of the above-mentioned learning users may be determined as a row identification set. Then, the row identifications in the above-mentioned row identification set may be sorted by a sorting algorithm to obtain a row identification sequence.

[0030] Step 4, construct a positive isomorphic graph matrix based on the above-mentioned correctly answered exercise data sequence, the above-mentioned first column identifier sequence and the above-mentioned row identifier sequence. Wherein, the above-mentioned positive isomorphic graph matrix can be an adjacency matrix that characterizes the above-mentioned positive isomorphic graph. First, the row identifiers in the above-mentioned row identifier sequence are used as rows, the first column identifiers in the above-mentioned first column identifier sequence are used as columns, and 0 is used as the element value to construct an initial positive isomorphic graph matrix. Wherein, the row sequence number corresponding to the row identifier in the initial positive isomorphic graph matrix can be the same as the sequence number of the row identifier in the row identifier sequence. The column sequence number corresponding to the first column identifier in the initial positive isomorphic graph matrix can be the same as the sequence number of the first column identifier in the first column identifier sequence. Then, for each correctly answered exercise data included in the above-mentioned correctly answered exercise data sequence, each element value in the above-mentioned initial positive isomorphic graph matrix that matches the above-mentioned correctly answered exercise data is updated to 1. Wherein, matching with the above-mentioned correctly answered exercise data can be: the row identifier and the first column identifier corresponding to the element value of the initial positive isomorphic graph matrix are respectively the same as the learning user identifier and the arbitrary exercise identifier corresponding to the above-mentioned correctly answered exercise data. Finally, the updated initial positive isomeric graph matrix is ​​determined as the positive isomeric graph matrix.

[0031] Step 5, based on the above-mentioned incorrectly answered exercise data sequence, the above-mentioned second column identifier sequence and the above-mentioned row identifier sequence, a negative isomorphic graph matrix is ​​constructed. Wherein, the above-mentioned negative isomorphic graph matrix can be an adjacency matrix that characterizes the above-mentioned negative isomorphic graph. First, the row identifiers in the above-mentioned row identifier sequence are used as rows, the second column identifiers in the above-mentioned second column identifier sequence are used as columns, and 0 is used as the element value to construct an initial negative isomorphic graph matrix. Wherein, the row sequence number corresponding to the row identifier in the initial negative isomorphic graph matrix can be the same as the sequence number of the row identifier in the row identifier sequence. The column sequence number corresponding to the second column identifier in the initial negative isomorphic graph matrix can be the same as the sequence number of the second column identifier in the second column identifier sequence. Then, for each incorrectly answered exercise data included in the above-mentioned incorrectly answered exercise data sequence, each element value in the above-mentioned initial negative isomorphic graph matrix that matches the above-mentioned incorrectly answered exercise data is updated to 1. Wherein, matching with the above-mentioned incorrectly answered exercise data can be: the row identifier and the second column identifier corresponding to the element value of the initial negative isomorphic graph matrix are respectively the same as the learning user identifier and any exercise identifier corresponding to the above-mentioned incorrectly answered exercise data. Finally, the updated initial negative heterogeneous graph matrix is ​​determined as the negative heterogeneous graph matrix.

[0032] Step six: determine the positive heterogeneous graph matrix and the negative heterogeneous graph matrix as user exercise heterogeneous graph data corresponding to each learning user.

[0033] The third step is to generate a hypergraph association matrix of the corresponding exercise concept hypergraph as exercise concept hypergraph data based on the above exercise concept association data sequence, taking the preset exercise identification in the preset exercise identification sequence as rows and the preset concept identification in the preset concept identification sequence as columns. The above preset exercise identification sequence can be a sequence of each preset concept identification. First, the preset exercise identification in the preset exercise identification sequence is taken as rows, the preset concept identification in the preset concept identification sequence is taken as columns, and 0 is taken as the element value to construct an initial hypergraph association matrix. Then, for each exercise concept association data in the above exercise concept association data sequence, each element value in the above initial hypergraph association matrix that matches the above exercise concept association data is updated to 1. Matching with the above exercise concept association data can be: the preset exercise identification and the preset concept identification corresponding to the element value of the initial hypergraph association matrix are respectively the same as any exercise identification and concept identification included in the above exercise concept association data. Finally, the updated initial hypergraph association matrix is ​​determined as the hypergraph association matrix.

[0034] The fourth step is to determine the set of exercise data pairs corresponding to the above-mentioned user exercise interaction data sequence. Each exercise data pair in the above-mentioned exercise data pair set may be data of two adjacent and ordered exercises that the user has answered successively. Each exercise data pair in the above-mentioned exercise data pair set may include a first exercise identifier and a second exercise identifier. The above-mentioned first exercise identifier may represent the exercise that the user answers first. The above-mentioned second exercise identifier may represent the next exercise that the user continues to answer after answering the exercise corresponding to the first exercise identifier. The preset exercise data extraction interface may be called to extract each exercise answer data sequence included in the above-mentioned user exercise interaction data sequence to obtain an exercise data pair set. The above-mentioned exercise data extraction interface may be encapsulated with an exercise data extraction method. The above-mentioned exercise data extraction method may be used to extract the identifiers of two adjacent and ordered exercises that the user has answered successively from the exercise answer data sequence as exercise data pairs.

[0035] Step 5: Select the exercise data pairs that meet the preset interaction times condition from the exercise data pair set to obtain an exercise data pair group. The preset interaction times condition may be that the exercise data pair appears more than 0 times in the answer record of each learning user.

[0036] The sixth step is to generate an exercise out-degree adjacency matrix and an exercise in-degree adjacency matrix based on the above exercise data pair group. The above exercise out-degree adjacency matrix can be an adjacency matrix that represents the above exercise directed transition graph, with out-degree nodes as rows and in-degree nodes as columns. The element value in the above exercise out-degree adjacency matrix can represent whether there is a directed edge from the corresponding out-degree node to the corresponding in-degree node. For example, when there is a directed edge from the out-degree node corresponding to the element value to the corresponding in-degree node, the element value can be 1, otherwise, the element value can be 0. The above exercise in-degree adjacency matrix can be an adjacency matrix that represents the above exercise directed transition graph, with in-degree nodes as rows and out-degree nodes as columns. The element value in the above exercise in-degree adjacency matrix can represent whether there is a directed edge from the corresponding out-degree node to the corresponding in-degree node. First, each first exercise identifier in the above exercise data pair group is determined as a first exercise identifier group, and each second exercise identifier in the above exercise data pair group is determined as a second exercise identifier group. Then, through the preset adjacency matrix construction interface, according to the above-mentioned first exercise identification group, the above-mentioned second exercise identification group and the above-mentioned exercise data pair group, the exercise out-degree adjacency matrix is ​​constructed. Among them, the above-mentioned adjacency matrix construction interface can be encapsulated with an adjacency matrix construction method. The above-mentioned adjacency matrix construction method can be used to take the first exercise identification group as the out-degree node group and the second exercise identification group as the in-degree node group, and construct the exercise out-degree adjacency matrix according to the exercise data pair group representing the relationship between the first exercise identification and the second exercise identification. Finally, the transposed matrix corresponding to the above-mentioned exercise out-degree adjacency matrix is ​​determined as the exercise in-degree adjacency matrix.

[0037] The seventh step is to determine the above-mentioned exercise out-degree adjacency matrix and the above-mentioned exercise in-degree adjacency matrix as exercise directed transformation graph data.

[0038] In practice, the out-degree adjacency matrix of an exercise represents the directed edge relationship from one exercise to another, and the in-degree adjacency matrix of an exercise constructed by the transposition operation represents the inverse relationship between exercises. The use of the transposed matrix can better describe the bidirectional learning path between exercises, including the order in which students answer questions.

[0039] Step 103, input the user exercise interaction data sequence, the user exercise heterogeneous graph data, the exercise concept hypergraph data and the exercise directed transition graph data into the pre-trained knowledge tracking model to obtain each user knowledge state vector.

[0040] In some embodiments, the execution subject may input the user exercise interaction data sequence, the user exercise heterogeneous graph data, the exercise concept hypergraph data and the exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector. Among them, the user knowledge state vector in each user knowledge state vector may correspond to the learning user in each learning user. The user knowledge state vector may characterize the mastery of each concept knowledge point by the corresponding learning user. Each component in the user knowledge state vector corresponds to a concept knowledge point one by one. The knowledge tracking model may include a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module. The sequence encoding module may be used to encode the exercise sequence data into a vector representation. The first graph embedding module may be used to perform graph embedding processing on the user exercise heterogeneous graph. The second graph embedding module may be used to perform graph embedding processing on the exercise concept hypergraph. The third graph embedding module may be used to perform graph embedding processing on the exercise directed transition graph. The interactive feature fusion module can be used to fuse the exercise sequence data and the graph embedding data and construct an initial knowledge state vector. The gated recurrent unit module can be used to update the initial knowledge state vector to obtain a current knowledge state vector.

[0041] Optionally, the above knowledge tracking model can be pre-trained by the following steps: The first step is to obtain a sample user exercise interaction data sequence, a sample exercise concept association data sequence and an initial knowledge tracking model. Among them, each sample user exercise interaction data in the above sample user exercise interaction data sequence may include a sample exercise answer data sequence and a sample exercise answer label. The above sample user exercise interaction data sequence may correspond to a sample learning user identification set one by one. The sample learning user identification in the above sample learning user identification set may be an identification of a learning user as a sample. The above sample user exercise interaction data sequence may be a user exercise interaction data sequence associated with the corresponding learning user. The above sample exercise answer label may be a correct answer probability matrix of the corresponding learning user when actually answering each exercise in the next time step. The rows of the correct answer probability matrix represent exercises, and the columns represent the probability of correct answers. The element values ​​in the correct answer probability matrix may represent the possibility of the learning user answering the corresponding exercise correctly when actually answering the corresponding exercise in the next time step. The above sample exercise concept association data sequence may be an exercise concept association data sequence as a sample. The above initial knowledge tracking model may be an untrained knowledge tracking model. The above-mentioned initial knowledge tracking model may include a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module.

[0042] The second step is to generate sample user exercise heterogeneous graph data, sample exercise concept hypergraph data and sample exercise directed transition graph data based on the sample user exercise interaction data sequence and the sample exercise concept association data sequence. Among them, the sample user exercise heterogeneous graph data, sample exercise concept hypergraph data and sample exercise directed transition graph data can be user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data as samples, respectively. Referring to the above step 102, based on the sample user exercise interaction data sequence and the sample exercise concept association data sequence, the sample user exercise heterogeneous graph data, sample exercise concept hypergraph data and sample exercise directed transition graph data are generated. The specific generation steps will not be repeated here.

[0043] The third step is to use the sample user exercise interaction data sequence as a training sample set, select training samples from the training sample set, and perform the following training steps: In the first sub-step, the sample exercise answer data sequence included in the above training samples is input into the sequence encoding module included in the initial knowledge tracking model to obtain an exercise answer data vector sequence. Among them, the exercise answer data vectors in the exercise answer data vector sequence correspond to the sample exercise answer data in the sample exercise answer data sequence one-to-one. The exercise answer data vectors in the exercise answer data vector sequence can represent the corresponding sample exercise answer data. The above sequence encoding module can be used to perform vector encoding representation on the sample exercise answer data in the sample exercise answer data sequence in sequence through the one-hot encoding method to obtain an exercise answer data vector sequence.

[0044] The second sub-step is to input the sample user exercise heterogeneous graph data into the first graph embedding module included in the initial knowledge tracking model to obtain a positive graph exercise embedding matrix, a negative graph exercise embedding matrix, a positive graph user embedding matrix and a negative graph user embedding matrix. Among them, the first graph embedding module can be composed of a lightweight graph convolutional network. The positive graph exercise embedding matrix can be composed of each positive graph exercise embedding vector. The positive graph exercise embedding vector in each positive graph exercise embedding vector can represent the exercises that are answered correctly. The negative graph exercise embedding matrix can be composed of each negative graph exercise embedding vector. The negative graph exercise embedding vector in each negative graph exercise embedding vector can represent the exercises that are answered incorrectly. The positive graph user embedding matrix can be composed of each positive graph user embedding vector. The positive graph user embedding vector in the positive graph user embedding vector can represent the users who answer the exercises correctly. The negative graph user embedding matrix can be composed of each negative graph user embedding vector. The negative graph user embedding vector in the negative graph user embedding vector can represent the users who answer the exercises incorrectly.

[0045] The third sub-step is to fuse the positive graph exercise embedding matrix and the negative graph exercise embedding matrix to obtain a first exercise embedding matrix. The first exercise embedding matrix can be composed of various first exercise embedding vectors. Each first exercise embedding vector can correspond to an exercise one-to-one. Each first exercise embedding vector can represent the relationship between the corresponding exercise and other exercises. The positive graph exercise embedding matrix and the negative graph exercise embedding matrix can be fused through a gating unit according to the following formula to obtain a first exercise embedding matrix: .

[0046] in, represents the embedding matrix of the first exercise. Represents the embedding matrix of the positive graph exercise. Represents the final embedding of the negative image. Represents the gate used to balance the importance of information learned from positive and negative heterogeneous graphs. Represents the weight parameter matrix. Indicates deviation. Represents the Hadamard product. Represents the ReLU activation function.

[0047] The fourth sub-step is to input the sample exercise concept hypergraph data into the second graph embedding module included in the initial knowledge tracking model to obtain a concept embedding matrix. The second graph embedding module may be composed of a hypergraph neural network. The concept embedding matrix may be composed of individual concept embedding vectors. The concept embedding vectors in the individual concept embedding vectors may represent the association between the corresponding exercise node and each concept knowledge point.

[0048] The fifth sub-step is to input the directed transformation graph data of the sample exercises into the third graph embedding module included in the initial knowledge tracking model to obtain the second exercise embedding matrix. The third graph embedding module may be composed of a directed graph neural network. The second exercise embedding matrix may be composed of individual second exercise embedding vectors. Each second exercise embedding vector may correspond to an exercise one-to-one. Each second exercise embedding vector may represent the directed relationship between the corresponding exercise and other exercises.

[0049] The sixth sub-step is to use the interactive feature fusion module included in the above-mentioned initial knowledge tracking model to generate the initial first knowledge state vector, the initial second knowledge state vector and the initial third knowledge state vector based on the above-mentioned exercise answer data vector sequence, the above-mentioned first exercise embedding matrix, the above-mentioned concept embedding matrix and the above-mentioned second exercise embedding matrix. Among them, the above-mentioned initial first knowledge state vector may correspond to the first exercise embedding matrix, representing the basic state of the learning user before contacting these exercises. The above-mentioned initial second knowledge state vector may correspond to the concept embedding matrix, representing the learning user's initial understanding or association with specific concepts. The above-mentioned initial third knowledge state vector may correspond to the second exercise embedding matrix, representing the learning user's initial mastery of the exercises.

[0050] In some optional implementations of some embodiments, the interactive feature fusion module may include a self-attention submodule. The self-attention submodule may be composed of an attention neural network. The attention neural network may be used to assign an attention weight representing relative importance to each element in the input sequence and perform weighted summation on these elements. The execution subject may generate an initial first knowledge state vector, an initial second knowledge state vector, and an initial third knowledge state vector based on the exercise answer data vector sequence, the first exercise embedding matrix, the concept embedding matrix, and the second exercise embedding matrix through the following substeps: Sub-step 1: for each exercise answer data vector in the above exercise answer data vector sequence, perform the following steps: The first step is to determine the first target exercise embedding vector, the second target exercise embedding vector and the third target exercise embedding vector corresponding to the exercise answer data vector based on the first exercise embedding matrix, the concept embedding matrix and the second exercise embedding matrix. Among them, the first target exercise embedding vector can be the first exercise embedding vector corresponding to the exercise answer data vector in the first exercise embedding matrix. The second target exercise embedding vector can be the concept embedding vector corresponding to the exercise answer data vector in the concept embedding matrix. The third target exercise embedding vector can be the second exercise embedding vector corresponding to the exercise answer data vector in the second exercise embedding matrix. First, the exercise identifier corresponding to the exercise answer data vector is determined as the target exercise identifier. Then, the first exercise embedding vector whose corresponding exercise identifier is the same as the target exercise identifier is selected from the first exercise embedding matrix as the first target exercise embedding vector. After that, the concept embedding vector whose corresponding exercise identifier is the same as the target exercise identifier is selected from the concept embedding matrix as the second target exercise embedding vector. Finally, a second exercise embedding vector whose corresponding exercise identifier is the same as the target exercise identifier is selected from the second exercise embedding matrix as the third target exercise embedding vector.

[0051] The second step is to generate the first information fusion feature value, the second information fusion feature value and the third information fusion feature based on the above-mentioned exercise answer data vector, the above-mentioned first target exercise embedding vector, the above-mentioned second target exercise embedding vector and the above-mentioned third target exercise embedding vector. Among them, the above-mentioned first information fusion feature value can represent the exercise answer data that integrates the user exercise heterogeneous graph embedding information. The above-mentioned second information fusion feature value can represent the exercise answer data that integrates the exercise concept graph embedding information. The above-mentioned third information fusion feature value can represent the exercise answer data that integrates the directed transformation graph embedding information. First, perform a dot multiplication operation on the above-mentioned exercise answer data vector and the above-mentioned first target exercise embedding vector to obtain the first information fusion feature value. Then, perform a dot multiplication operation on the above-mentioned exercise answer data vector and the above-mentioned second target exercise embedding vector to obtain the second information fusion feature value. Finally, perform a dot multiplication operation on the above-mentioned exercise answer data vector and the above-mentioned third target exercise embedding vector to obtain the third information fusion feature value.

[0052] Sub-step 2: Generate a first information fusion vector based on each first information fusion feature value corresponding to the above exercise answer data vector sequence. The above first information fusion vector can represent the exercise answer data sequence that integrates the embedded information of the user exercise heterogeneous graph. The first information fusion feature value corresponding to each exercise answer data vector in the above exercise answer data vector sequence can be used as the component at the corresponding position in the vector to obtain the first information fusion vector.

[0053] Sub-step three, based on each second information fusion eigenvalue corresponding to the above exercise answer data vector sequence, generate a second information fusion vector. The above second information fusion vector can represent the exercise answer data sequence that integrates the exercise concept hypergraph embedding information. The second information fusion eigenvalue corresponding to each exercise answer data vector in the above exercise answer data vector sequence can be used as the component at the corresponding position in the vector to obtain the second information fusion vector.

[0054] Sub-step 4: Generate a third information fusion vector based on each third information fusion eigenvalue corresponding to the above exercise answer data vector sequence. The above third information fusion vector can represent the exercise answer data sequence that incorporates the directed transition graph embedded information. The third information fusion eigenvalue corresponding to each exercise answer data vector in the above exercise answer data vector sequence can be used as a component at a corresponding position in the vector to obtain the third information fusion vector.

[0055] Sub-step five: input the first information fusion vector into the self-attention sub-module to obtain an initial first knowledge state vector.

[0056] Sub-step six, inputting the second information fusion vector into the self-attention sub-module to obtain an initial second knowledge state vector.

[0057] Sub-step seven, inputting the third information fusion vector into the self-attention sub-module to obtain an initial third knowledge state vector.

[0058] In practice, the above-mentioned execution entity can introduce the complex relationship between the positions in the sequence by inputting the exercise vector sequence that integrates the graph embedding information into the above-mentioned self-attention sub-module, simulate the "inertia" of students in the process of solving problems, and capture the internal information in the graph fusion features, so as to better model the changes in students' knowledge status.

[0059] The seventh sub-step is to use the gated recurrent unit module included in the above-mentioned initial knowledge tracking model to update the above-mentioned initial first knowledge state vector, the above-mentioned initial second knowledge state vector and the above-mentioned initial third knowledge state vector to obtain the first knowledge state vector, the second knowledge state vector and the third knowledge state vector. Among them, the above-mentioned first knowledge state vector can represent the mastery of each concept knowledge point corresponding to the exercise answer data sequence that integrates the user exercise heterogeneous graph embedding information. The above-mentioned second knowledge state vector can represent the mastery of each concept knowledge point corresponding to the exercise answer data sequence that integrates the exercise concept graph embedding information. The above-mentioned third knowledge state vector can represent the mastery of each concept knowledge point corresponding to the exercise answer sequence that integrates the directed transition graph embedding information. The gated recurrent unit module included in the above-mentioned initial knowledge tracking model can be used to update the above-mentioned first knowledge state vector, the above-mentioned second knowledge state vector and the above-mentioned initial third knowledge state vector according to the following formula group to obtain the first knowledge state vector, the second knowledge state vector and the third knowledge state vector: .

[0060] in, Represents the current time step. Indicates the previous time step. Each time step corresponds to the data of the exercise to be answered. Represents the knowledge state vector corresponding to the previous time step. Represents the knowledge state vector corresponding to the current time step. Represents the output of the self-attention submodule. Represents the reset gate, which is used to determine the state that should be known The amount of information forgotten. represents the update gate, which is used to determine the new addition to the knowledge state The amount of information in Represents a storage unit, used to store information to be added to the knowledge state Information in. Can be balanced and . , , is the transformation matrix, , , Represents the bias term to be learned. Represents the sigmoid activation function. Represents the tanh activation function.

[0061] The eighth sub-step is to map the first knowledge state vector, the second knowledge state vector and the third knowledge state vector to obtain the first correct answer probability matrix, the second correct answer probability matrix and the third correct answer probability matrix. Among them, the first correct answer probability matrix can be the probability matrix of the correct answer of the learning user corresponding to the first knowledge state vector and predicted when answering the exercises in the next time step. The second correct answer probability matrix can be the probability matrix of the correct answer of the learning user corresponding to the second knowledge state vector and predicted when answering the exercises in the next time step. The third correct answer probability matrix can be the probability matrix of the correct answer of the learning user corresponding to the third knowledge state vector and predicted when answering the exercises in the next time step. The first knowledge state vector, the second knowledge state vector and the third knowledge state vector can be mapped by the Sigmoid activation function to obtain the first correct answer probability matrix, the second correct answer probability matrix and the third correct answer probability matrix.

[0062] The ninth sub-step is to determine the first loss value, the second loss value and the third loss value based on the correct answer labels of the sample exercises corresponding to the above training samples, the first correct answer probability matrix, the second correct answer probability matrix and the third correct answer probability matrix. Among them, the first loss value can characterize the difference between the correct answer labels of the sample exercises and the first correct answer probability matrix. The second loss value can characterize the difference between the correct answer labels of the sample exercises and the second correct answer probability matrix. The third loss value can characterize the difference between the correct answer labels of the sample exercises and the third correct answer probability matrix. The first loss value, the second loss value and the third loss value can be determined based on the correct answer labels of the sample exercises corresponding to the above training samples, the first correct answer probability matrix, the second correct answer probability matrix and the third correct answer probability matrix through the cross entropy loss function.

[0063] The tenth sub-step is to determine a fourth loss value based on the sample exercise answer data sequence included in the training sample, the positive image exercise embedding matrix, the negative image exercise embedding matrix, the positive image user embedding matrix, and the negative image user embedding matrix. The fourth loss value can represent the difference between the exercises answered correctly and the exercises answered incorrectly.

[0064] In some optional implementations of some embodiments, each sample exercise answer data in the sample exercise answer data sequence may include a sample exercise identifier. The sample exercise identifier may be an exercise identifier corresponding to the exercise included in the sample for training the model. The execution subject may determine the fourth loss value based on the sample exercise answer data sequence included in the training sample, the positive image exercise embedding matrix, the negative image exercise embedding matrix, the positive image user embedding matrix, and the negative image user embedding matrix through the following sub-steps: Sub-step one, based on the positive image user embedding matrix and the negative image user embedding matrix, determine the sample positive image user embedding vector and the sample negative image user embedding vector corresponding to the training sample. First, determine the sample learning user identifier corresponding to the training sample as the target sample learning user identifier. Then, select the positive image user embedding vector whose corresponding learning user identifier is the same as the target sample learning user identifier from the positive image user embedding matrix as the sample positive image user embedding vector. Finally, select the negative image user embedding vector whose corresponding learning user identifier is the same as the target sample learning user identifier from the negative image user embedding matrix as the sample negative image user embedding vector.

[0065] Sub-step 2: for each sample exercise identifier included in the sample exercise answer data sequence in the training sample, the following steps may be performed: The first step is to determine the sample positive image exercise embedding vector and the sample negative image exercise embedding vector corresponding to the sample exercise identifier based on the positive image exercise embedding matrix and the negative image exercise embedding matrix. First, select the positive image exercise embedding vector whose corresponding exercise identifier is the same as the sample exercise identifier from the positive image exercise embedding matrix as the sample positive image exercise embedding vector. Then, select the negative image exercise embedding vector whose corresponding exercise identifier is the same as the sample exercise identifier from the negative image exercise embedding matrix as the sample negative image exercise embedding vector.

[0066] In the second step, the dot product result of the above sample positive image user embedding vector and the above sample positive image exercise embedding vector is determined as the correct answer prediction score. Among them, the correct answer prediction score represents the possibility that the corresponding learning user can correctly answer the corresponding exercise. The higher the correct answer prediction score, the greater the possibility that the model believes that the corresponding learning user can correctly answer the question.

[0067] The third step is to determine the dot product result of the above sample negative graph user embedding vector and the above sample negative graph exercise embedding vector as the wrong answer prediction score. Among them, the wrong answer prediction score represents the possibility that the corresponding learning user cannot correctly answer the corresponding exercise. The lower the wrong answer prediction score, the greater the possibility that the model believes that the corresponding learning user cannot correctly answer the question.

[0068] The fourth step is to determine the sample exercise identification, the correct answer prediction score and the incorrect answer prediction score as exercise score information.

[0069] Sub-step three: Generate a fourth loss value based on the obtained score information of each exercise. The fourth loss value can be generated based on the obtained score information of each exercise by the following loss function: .

[0070] in, Represents the fourth loss value. Indicates a sequence number. Represents the total number of sample exercises. Represents the predicted score for the questions that the user answered correctly. Indicates the predicted scores for questions that the user answered incorrectly or did not answer. Indicates The predicted score when the user answers the question correctly. Indicates The predicted score for a question when the user answers it incorrectly or does not answer it. A small positive number used to prevent numerical instability when taking logarithms. Indicates that Restricted to the interval Inside.

[0071] In practice, the above-mentioned execution entity first constructs a user exercise heterogeneous graph, and then learns the user's thinking tendency towards exercises in the positive heterogeneous graph and negative heterogeneous graph included in the user exercise heterogeneous graph through a lightweight graph convolutional network. Finally, the model parameters are trained through the above-mentioned loss function for graph contrast learning, so that the predicted score of the user's correct answer to the question is higher than the predicted score of the user's incorrect answer or no answer, thereby highlighting the user's tendency to answer the exercises correctly, thereby simulating the phenomenon that students master familiar content faster during the learning process.

[0072] In the eleventh sub-step, a total loss value is generated based on the first loss value, the second loss value, the third loss value and the fourth loss value. First, the product of the fourth loss value and the preset weight value is determined as the adjusted fourth loss value. The preset weight value may be a pre-set weight value, which may be used to control the contribution of the first loss value, the second loss value, the third loss value and the fourth loss value. Then, the sum of the first loss value, the second loss value, the third loss value and the adjusted fourth loss value is determined as the total loss value.

[0073] In the twelfth sub-step, in response to determining that the total loss value is less than a preset threshold, the trained initial knowledge tracking model is determined as the knowledge tracking model, wherein the preset threshold may be a preset upper limit of the loss value.

[0074] Optionally, the execution subject may also adjust relevant parameters in the initial knowledge tracking model in response to determining that the total loss value is not less than the preset threshold, and reselect training samples from the training sample set, use the adjusted initial knowledge tracking model as the initial knowledge tracking model, and perform the training step again. The parameters in the initial knowledge tracking model may be adjusted using the Adam optimizer.

[0075] Step 104: for each of the learning users, perform the following steps: Step 1041 , performing detection processing on each component in the user knowledge state vector corresponding to the learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence.

[0076] In some embodiments, the execution subject may detect and process each component in the user knowledge state vector corresponding to the learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence. The first knowledge state component information in the first knowledge state component information sequence may represent a concept knowledge point with a low degree of mastery corresponding to the learning user. The second knowledge state component information in the second knowledge state component information sequence may represent a concept knowledge point with a high degree of mastery corresponding to the learning user. For each component in the target user knowledge state vector, perform the following steps: The first sub-step is to determine the position number corresponding to the component in the target user knowledge state vector as the target number. The preset mastering degree value may be a preset value representing the mastering degree of the concept knowledge point.

[0077] In the second sub-step, a concept identifier having the same serial number as the target serial number is selected from the concept identifier sequence corresponding to the exercise concept hypergraph data as the concept identifier corresponding to the component.

[0078] The third sub-step is, in response to determining that the above component is less than a preset mastery level value, determining the value of the above component and the corresponding concept identifier as first knowledge state component information.

[0079] Optionally, the execution subject may also determine the value of the component and the corresponding concept identifier as the second knowledge state component information in response to determining that the component is not less than the preset mastery level value.

[0080] Step 1042: Generate a second knowledge state post-concept information set based on the second knowledge state component information sequence and the preset concept learning path information set.

[0081] In some embodiments, the execution subject may generate a second knowledge state post-concept information set based on the second knowledge state component information sequence and the preset concept learning path information set. The preset concept learning path information in the preset concept learning path information set may be pre-set information that characterizes the pre-relationship between two concept knowledge points. The preset concept learning path information in the preset concept learning path information set may include a pre-concept identifier and a post-concept identifier. The pre-concept identifier may be an identifier of another concept knowledge point that needs to be mastered in advance before a concept. The post-concept identifier may be an identifier of another concept knowledge point that needs to be mastered after mastering a concept. The second knowledge state post-concept information group may be a set of non-repetitive post-concept identifiers corresponding to each target concept identifier. The target concept identifier in each target concept identifier may be a concept identifier included in the second knowledge state component information in the second knowledge state component information sequence. Specifically, the following steps may be performed: The first step is to select, for each concept identifier included in the above-mentioned second knowledge state component information sequence, the preset concept learning path information including the preceding concept identifier which is the same as the above-mentioned concept identifier from the above-mentioned preset concept learning path information set to obtain a preset concept learning path information group, and determine each post-concept identifier included in the above-mentioned preset concept learning path information group as a post-concept identifier group.

[0082] In the second step, each post-concept identifier in each post-concept identifier group is de-redundanted to obtain a post-concept identifier set after de-redundancy as the post-concept information set of the second knowledge state, wherein the post-concept identifiers after de-redundancy in the post-concept identifier set after de-redundancy can be non-repeated post-concept identifiers.

[0083] Step 1043, based on the exercise concept association data sequence, the first knowledge state component information sequence and the second knowledge state post-concept information set, generate user exercise recommendation information, and store the user exercise recommendation information to the cloud server for learning users to obtain.

[0084] In some embodiments, the execution subject may generate user exercise recommendation information based on the exercise concept association data sequence, the first knowledge state component information sequence and the second knowledge state post-concept information group in various ways, and store the user exercise recommendation information in the cloud server for the learning user to obtain. The user exercise recommendation information may be information about the next exercise to be answered to be pushed to the user.

[0085] Optionally, after receiving a network request from a learning user to obtain recommended user exercise information, the cloud server pushes the recommended user exercise information to the corresponding learning user for practice.

[0086] In some optional implementations of some embodiments, the execution subject may generate user exercise recommendation information based on the exercise concept association data sequence, the first knowledge state component information sequence and the second knowledge state post-concept information set through the following steps: The first step is to determine the first to-be-pushed exercise identification group corresponding to each first knowledge state component information in the first knowledge state component information sequence based on the exercise concept association data sequence, and obtain the first to-be-pushed exercise identification group set. For each first knowledge state component information in the first knowledge state component information sequence, the following steps can be performed: In the first sub-step, the exercise concept associated data including the same concept identifier as the concept identifier included in the first knowledge state component information is selected from the exercise concept associated data sequence as the first target exercise concept associated data.

[0087] The second sub-step is to determine the exercise identification group included in the first target exercise concept association data as the first exercise identification group to be pushed.

[0088] The second step is to determine the second to-be-pushed exercise identification group corresponding to each second knowledge state post-position concept information in the second knowledge state post-position concept information set based on the exercise concept association data sequence, and obtain the second to-be-pushed exercise identification group set. For each second knowledge state post-position concept information in the second knowledge state post-position concept information set, the following steps can be performed: In the first sub-step, the exercise concept associated data including the same concept identifier as the second knowledge state post-concept information is selected from the exercise concept associated data sequence as the second target exercise concept associated data.

[0089] The second sub-step is to determine the exercise identification group included in the second target exercise concept association data as the second exercise identification group to be pushed.

[0090] The third step is to perform redundancy processing on each first exercise identification to be pushed in each obtained first exercise identification group to obtain a first redundancy-free exercise identification set. The first redundancy-free exercise identification set can be a collection of non-repeated first exercise identifications to be pushed.

[0091] The fourth step is to perform redundancy processing on each second exercise identification to be pushed in the second exercise identification group to obtain a second redundancy-free exercise identification set. The second redundancy-free exercise identification set can be a collection of non-repeated second exercise identifications to be pushed.

[0092] Step 5: For each first de-redundant exercise identifier in the first de-redundant exercise identifier set, perform the following steps: In the first sub-step, the exercise difficulty information matching the first redundant exercise identification is selected from the preset exercise difficulty information set as the target exercise difficulty information. The exercise difficulty information in the above exercise difficulty information set may include an exercise identification and a difficulty coefficient. The difficulty coefficient may represent the difficulty of answering the exercise correctly. The exercise difficulty information in the above exercise difficulty information set may be generated based on the ratio between the number of learning users who have not answered the corresponding exercise correctly and the total number of learning users.

[0093] The second sub-step is to determine the graded mark corresponding to the first redundant exercise mark based on the target exercise difficulty information. The graded mark may indicate the difficulty interval where the exercise difficulty is located. First, in response to determining that the difficulty coefficient included in the target exercise difficulty information is less than or equal to the first preset division value, the preset primary mark is determined as the graded mark. The first preset division value may be a difficulty coefficient preset to separate the low difficulty interval and the medium difficulty interval. The preset primary mark may indicate that the exercise difficulty of the corresponding exercise is located in the low difficulty interval. Then, in response to determining that the difficulty coefficient included in the target exercise difficulty information is greater than the first preset division value and less than or equal to the second preset division value, the preset intermediate mark is determined as the graded mark. The second preset division value may be a difficulty coefficient preset to separate the medium difficulty interval and the high difficulty interval. The preset intermediate mark may indicate that the exercise difficulty of the corresponding exercise is located in the medium difficulty interval. Finally, in response to determining that the difficulty coefficient included in the target exercise difficulty information is greater than the second preset division value, the preset advanced mark is determined as the graded mark. The above-mentioned preset advanced mark can indicate that the difficulty level of the corresponding exercise is in a high difficulty range.

[0094] The sixth step is to classify the first de-redundant exercise identifications in the above-mentioned first de-redundant exercise identification set based on the hierarchical identification corresponding to the first de-redundant exercise identification, and obtain a first-level exercise identification set, a second-level exercise identification set and a third-level exercise identification set. Among them, the above-mentioned first-level exercise identification set can be a collection of first de-redundant exercise identifications whose hierarchical identification is the same as the preset primary identification. The above-mentioned second-level exercise identification set can be a collection of first de-redundant exercise identifications whose hierarchical identification is the same as the preset intermediate identification. The above-mentioned third-level exercise identification set can be a collection of first de-redundant exercise identifications whose hierarchical identification is the same as the preset advanced identification.

[0095] The seventh step is to determine the first recommended exercise identification set based on the preset exercise extraction ratio information group, the above-mentioned first-level exercise identification set, the above-mentioned second-level exercise identification set and the above-mentioned third-level exercise identification set. Among them, the preset exercise extraction ratio information in the above-mentioned preset exercise extraction ratio information group may include a grading identification and an extraction ratio. The above-mentioned extraction ratio may be the ratio when extracting exercises from each exercise with a corresponding grading identification. Through a preset exercise extraction interface, based on the preset exercise extraction ratio information group, the above-mentioned first-level exercise identification set, the above-mentioned second-level exercise identification set and the above-mentioned third-level exercise identification set may be extracted respectively, and each first-level exercise identification, each second-level exercise identification and each third-level exercise identification extracted may be determined as the first recommended exercise identification to obtain the first recommended exercise identification set.

[0096] Step 8: Delete each first recommended exercise identification that matches the first recommended exercise identification set in the second de-redundant exercise identification set to obtain the deleted second de-redundant exercise identification set. Matching the first recommended exercise identification set may be that the second de-redundant exercise identification in the second de-redundant exercise identification set is the same as any first recommended exercise identification in the first recommended exercise identification set.

[0097] The ninth step is to select a preset number of second de-redundant exercise identifications after deletion from the above-mentioned second de-redundant exercise identification set based on the preset extended exercise extraction ratio, and determine each selected second de-redundant exercise identification after deletion as the second recommended exercise identification set. The preset extended exercise extraction ratio may be a preset ratio required for extracting each second de-redundant exercise identification after deletion. The preset number may be the product of the number of elements in the above-mentioned second de-redundant exercise identification set after deletion and the preset extended exercise extraction ratio.

[0098] The tenth step is to generate user exercise recommendation information based on the preset exercise database, the first recommended exercise identification set and the second recommended exercise identification set. The exercise database may be a database for storing exercise detail information. The exercise detail information may include but is not limited to exercise identification, exercise title, question type, and options. First, obtain each exercise detail information corresponding to the first recommended exercise identification set and each exercise detail information corresponding to the second recommended exercise identification set from the exercise database. Then, the learning user identification of the learning user and the obtained each exercise detail information are determined as the user exercise recommendation information.

[0099] The above steps 1041 to 1043 and their related contents, as an inventive point of an embodiment of the present disclosure, solve the second technical problem mentioned in the background technology, "it is difficult to push the recommended user exercises information with a higher degree of adaptability to user needs to the learning user". The reasons for the difficulty in pushing the recommended user exercises information with a higher degree of adaptability to user needs to the learning user are often as follows: since the exercises recommended according to the knowledge tracking results can only cover the knowledge points learned in the historical period, the influence of the learner's learning path on the user's learning needs is ignored, so that the generated recommended user exercises information has a low degree of adaptability to the user's needs. If the above problem is solved, the effect of pushing the recommended user exercises information with a higher degree of adaptability to the user's needs to the learning user can be achieved. In order to achieve this effect, first, each component in the user knowledge state vector corresponding to the above learning user is detected and processed to obtain a first knowledge state component information sequence and a second knowledge state component information sequence. In this way, each concept knowledge point with a lower degree of mastery and each concept knowledge point with a higher degree of mastery can be determined. Then, based on the above second knowledge state component information sequence and the preset concept learning path information set, a second knowledge state post-concept information set is generated. Therefore, it is possible to determine the various concept knowledge points that the learning user needs to practice after mastering the various concept knowledge points at a higher level based on the learning path. Finally, based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, generate user exercise recommendation information, and store the above-mentioned user exercise recommendation information to the cloud server for the above-mentioned learning user to obtain. In this way, user exercise recommendation information with a higher degree of adaptability to user needs can be generated for user learning. In addition, because, when generating user exercise recommendation information based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, exercises of different difficulty levels are extracted according to the exercise difficulty information, thereby improving the discrimination of exercises for subsequent evaluation of learners' levels.

[0100] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the user exercise recommendation information generation method of some embodiments of the present disclosure, the user exercise recommendation information with a higher degree of adaptability to user needs can be pushed to the learning user. Specifically, the reason for pushing the user exercise recommendation information with a higher degree of adaptability to user needs to the learning user is that: due to the knowledge tracking technology based on the relationship between exercises and concept knowledge points, the center is exercises and concept knowledge points, and the learning user's preference for exercises is ignored, so that the generated user exercise recommendation information has a low degree of adaptability to user needs, thereby making it difficult to push the user exercise recommendation information with a higher degree of adaptability to user needs to the learning user. Based on this, the user exercise recommendation information generation method of some embodiments of the present disclosure, first, obtains the user exercise interaction data sequence and the exercise concept association data sequence corresponding to each learning user. Among them, each user exercise interaction data in the above-mentioned user exercise interaction data sequence includes an exercise answer data sequence. Thus, the historical exercise answering behavior of each learning user and the association relationship between exercises and concept knowledge points can be obtained. Secondly, based on the various exercise answer data sequences and the exercise concept association data sequences included in the above user exercise interaction data sequence, user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data are generated. Thus, high-order relationships between learning users and exercises, between exercises and concept knowledge points, and between exercises can be obtained. Then, the above user exercise interaction data sequence, the above user exercise heterogeneous graph data, the above exercise concept hypergraph data and the above exercise directed transition graph data are input into the pre-trained knowledge tracking model to obtain each user knowledge state vector. Among them, the user knowledge state vector in the above each user knowledge state vector corresponds to the learning user in the above each learning user, and the above knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module. Thus, through the knowledge tracking model, according to the connection between the learning user, the exercise and the concept, a knowledge state vector representing the user's knowledge mastery degree can be constructed for each learning user. Finally, for each of the above-mentioned learning users, the following steps are performed: each component in the user knowledge state vector corresponding to the above-mentioned learning user is detected and processed to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; based on the above-mentioned second knowledge state component information sequence and the preset concept learning path information set, a second knowledge state post-concept information set is generated; based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, user exercise recommendation information is generated, and the above-mentioned user exercise recommendation information is stored in the cloud server for the above-mentioned learning user to obtain. In this way, for each learning user, exercises related to the concept knowledge points that have not been mastered can be pushed to the learning user.Therefore, the method for generating user exercise recommendation information in some embodiments of the present disclosure can introduce students' preference information for exercises into the knowledge tracking model by inputting the user exercise heterogeneous graph data representing the high-order relationship between students and exercises into the knowledge tracking model. Furthermore, by integrating the relationship between students, exercises and concepts, a more accurate knowledge state can be generated for learning users. Thus, user exercise recommendation information that is more suitable for user needs can be generated. Furthermore, the number of recommendations for exercises with low adaptability can be reduced.

[0101] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a device for generating user exercise recommendation information. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the user exercise recommendation information generating device 200 can be specifically applied to various electronic devices.

[0102] like Figure 2As shown, some embodiments of the user exercise recommendation information generation device 200 include: an acquisition unit 201, a generation unit 202, an input unit 203 and an execution unit 204. The acquisition unit 201 is configured to acquire a user exercise interaction data sequence and an exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the above user exercise interaction data sequence includes an exercise answer data sequence; the generation unit 202 is configured to generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data based on each exercise answer data sequence and the above exercise concept association data sequence included in the above user exercise interaction data sequence; the input unit 203 is configured to input the above user exercise interaction data sequence, the above user exercise heterogeneous graph data, the above exercise concept hypergraph data and the above exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in the above each user knowledge state vector is correlated with the learning user in the above each learning user. Corresponding to the user, the above-mentioned knowledge tracking model may include a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; the execution unit 204 is configured to perform the following steps for each of the above-mentioned learning users: detect and process each component in the user knowledge state vector corresponding to the above-mentioned learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; generate a second knowledge state post-concept information set based on the above-mentioned second knowledge state component information sequence and a preset concept learning path information set; generate user exercise recommendation information based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, and store the above-mentioned user exercise recommendation information to the cloud server for the above-mentioned learning user to obtain.

[0103] It can be understood that the various units recorded in the user exercise recommendation information generating device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the user exercise recommendation information generating device 200 and the units included therein, and will not be described in detail here.

[0104] Further references Figure 3 , which shows a structural schematic diagram of an electronic device 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0105] like Figure 3As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0106] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0107] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0108] It should be noted that the computer-readable medium mentioned above in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0109] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0110] The above-mentioned computer-readable medium may be included in the above-mentioned device; or it may exist independently and not be installed in the electronic device. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: obtains the user exercise interaction data sequence and the exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the above-mentioned user exercise interaction data sequence includes an exercise answer data sequence; based on each exercise answer data sequence and the above-mentioned exercise concept association data sequence included in the above-mentioned user exercise interaction data sequence, generates user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data; inputs the above-mentioned user exercise interaction data sequence, the above-mentioned user exercise heterogeneous graph data, the above-mentioned exercise concept hypergraph data and the above-mentioned exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in the above-mentioned each user knowledge state vector is consistent with the above-mentioned user exercise interaction data sequence. Corresponding to the learning users among the above-mentioned learning users, the above-mentioned knowledge tracking model may include a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; for each learning user among the above-mentioned learning users, the following steps are performed: each component in the user knowledge state vector corresponding to the above-mentioned learning user is detected and processed to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; based on the above-mentioned second knowledge state component information sequence and the preset concept learning path information set, a second knowledge state post-concept information set is generated; based on the above-mentioned exercise concept association data sequence, the above-mentioned first knowledge state component information sequence and the above-mentioned second knowledge state post-concept information set, user exercise recommendation information is generated, and the above-mentioned user exercise recommendation information is stored in the cloud server for the above-mentioned learning user to obtain.

[0111] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0112] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0113] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be arranged in a processor, for example, may be described as: a processor comprising: an acquisition unit, a generation unit, an input unit, and an execution unit. The names of these units do not constitute limitations on the units themselves in certain circumstances, for example, the acquisition unit may also be described as "a unit for acquiring user exercise interaction data sequences and exercise concept association data sequences corresponding to each learning user".

[0114] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0115] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for generating user exercise recommendation information, comprising: Acquire a user exercise interaction data sequence and an exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the user exercise interaction data sequence includes an exercise answer data sequence; Generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transformation graph data based on each exercise answer data sequence and the exercise concept association data sequence included in the user exercise interaction data sequence; Input the user exercise interaction data sequence, the user exercise heterogeneous graph data, the exercise concept hypergraph data and the exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in each user knowledge state vector corresponds to a learning user in each learning user, and the knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; For each of the learning users, perform the following steps: Performing detection processing on each component in the user knowledge state vector corresponding to the learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; Generate a second knowledge state post-concept information set based on the second knowledge state component information sequence and the preset concept learning path information set; Based on the exercise concept association data sequence, the first knowledge state component information sequence and the second knowledge state post-concept information set, user exercise recommendation information is generated, and the user exercise recommendation information is stored in a cloud server for the learning user to obtain.

2. The method according to claim 1, wherein: The generating of user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transformation graph data based on each exercise answer data sequence and the exercise concept association data sequence included in the user exercise interaction data sequence comprises: Classify and process each exercise answer data included in the user exercise interaction data sequence to obtain a correct answer exercise data sequence and an incorrect answer exercise data sequence corresponding to each learning user; Based on the correctly answered exercise data sequence and the incorrectly answered exercise data sequence, generating user exercise heterogeneous graph data corresponding to each learning user; Taking the preset exercise identifiers in the preset exercise identifier sequence as rows and the preset concept identifiers in the preset concept identifier sequence as columns, based on the exercise concept association data sequence, a hypergraph association matrix corresponding to the exercise concept hypergraph is generated as exercise concept hypergraph data, wherein the exercise concept hypergraph is a hypergraph with exercises as nodes and concepts associated with exercises as hyperedges; Determine a set of exercise data pairs corresponding to the user exercise interaction data sequence, wherein each exercise data pair in the set of exercise data pairs is two adjacent and ordered exercise data that the user has answered successively; Selecting exercise data pairs that meet a preset interaction number condition from the exercise data pair set to obtain an exercise data pair group, wherein the preset interaction number condition is that the number of times the exercise data pair appears in the answer record of each learning user is greater than 0; Based on the exercise data pair, generating an exercise out-degree adjacency matrix and an exercise in-degree adjacency matrix; The exercise out-degree adjacency matrix and the exercise in-degree adjacency matrix are determined as exercise directed transition graph data.

3. The method according to claim 2, wherein: The generating of user exercise heterogeneous graph data corresponding to each learning user based on the correctly answered exercise data sequence and the incorrectly answered exercise data sequence includes: Determine the first column identification sequence corresponding to the correctly answered exercise data sequence; Determine a second column identification sequence corresponding to the incorrectly answered exercise data sequence; Determine the row identification sequence corresponding to each learning user; Constructing a positive heterogeneous graph matrix based on the correctly answered exercise data sequence, the first column identification sequence and the row identification sequence; Constructing a negative heterogeneous graph matrix based on the incorrectly answered exercise data sequence, the second column identification sequence and the row identification sequence; The positive heterogeneous graph matrix and the negative heterogeneous graph matrix are determined as user exercise heterogeneous graph data corresponding to each learning user.

4. The method according to claim 1, wherein: The knowledge tracking model is pre-trained by the following steps: Acquire a sample user exercise interaction data sequence, a sample exercise concept association data sequence and an initial knowledge tracking model, wherein each sample user exercise interaction data in the sample user exercise interaction data sequence includes a sample exercise answer data sequence and a sample exercise answer label, the sample user exercise interaction data sequence corresponds to a sample learning user identification set, the initial knowledge tracking model is an untrained knowledge tracking model, and the initial knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interaction feature fusion module and a gated recurrent unit module; Generate sample user exercise heterogeneous graph data, sample exercise concept hypergraph data and sample exercise directed transition graph data based on each sample exercise answer data sequence and the sample exercise concept association data sequence included in the sample user exercise interaction data sequence; The sample user exercise interaction data sequence is used as a training sample set, training samples are selected from the training sample set, and the following training steps are performed: Inputting the sample exercise answer data sequence included in the training sample into the sequence encoding module included in the initial knowledge tracking model to obtain the exercise answer data vector sequence; Inputting the sample user-question heterogeneous graph data into the first graph embedding module included in the initial knowledge tracking model to obtain a positive graph-question embedding matrix, a negative graph-question embedding matrix, a positive graph-user embedding matrix, and a negative graph-user embedding matrix; Fusing the positive graph exercise embedding matrix and the negative graph exercise embedding matrix to obtain a first exercise embedding matrix; Inputting the sample exercise concept hypergraph data into the second graph embedding module included in the initial knowledge tracking model to obtain a concept embedding matrix; Inputting the sample exercise directed transformation graph data into the third graph embedding module included in the initial knowledge tracking model to obtain a second exercise embedding matrix; Using the interactive feature fusion module included in the initial knowledge tracking model, based on the exercise answer data vector sequence, the first exercise embedding matrix, the concept embedding matrix and the second exercise embedding matrix, an initial first knowledge state vector, an initial second knowledge state vector and an initial third knowledge state vector are generated; Using the gated recurrent unit module included in the initial knowledge tracking model, the initial first knowledge state vector, the initial second knowledge state vector and the initial third knowledge state vector are updated to obtain a first knowledge state vector, a second knowledge state vector and a third knowledge state vector; Mapping the first knowledge state vector, the second knowledge state vector, and the third knowledge state vector to obtain a first correct answer probability matrix, a second correct answer probability matrix, and a third correct answer probability matrix; Determine a first loss value, a second loss value, and a third loss value based on the correct answer labels of the sample exercises corresponding to the training samples, the first correct answer probability matrix, the second correct answer probability matrix, and the third correct answer probability matrix; Determine a fourth loss value based on the sample exercise answer data sequence included in the training sample, the positive image exercise embedding matrix, the negative image exercise embedding matrix, the positive image user embedding matrix, and the negative image user embedding matrix; generating a total loss value based on the first loss value, the second loss value, the third loss value, and the fourth loss value; In response to determining that the total loss value is less than a preset threshold, the trained initial knowledge tracking model is determined as the knowledge tracking model.

5. The method according to claim 4, wherein: The interactive feature fusion module includes a self-attention submodule; and the generating of an initial first knowledge state vector, an initial second knowledge state vector and an initial third knowledge state vector based on the exercise answer data vector sequence, the first exercise embedding matrix, the concept embedding matrix and the second exercise embedding matrix includes: For each exercise answer data vector in the exercise answer data vector sequence, the following steps are performed: Determine a first target exercise embedding vector, a second target exercise embedding vector, and a third target exercise embedding vector corresponding to the exercise answer data vector based on the first exercise embedding matrix, the concept embedding matrix, and the second exercise embedding matrix; Generate a first information fusion feature value, a second information fusion feature value, and a third information fusion feature value based on the exercise answer data vector, the first target exercise embedding vector, the second target exercise embedding vector, and the third target exercise embedding vector; Generate a first information fusion vector based on each first information fusion feature value corresponding to the exercise answer data vector sequence; Generate a second information fusion vector based on each second information fusion feature value corresponding to the exercise answer data vector sequence; Generate a third information fusion vector based on each third information fusion feature value corresponding to the exercise answer data vector sequence; Inputting the first information fusion vector into the self-attention submodule to obtain an initial first knowledge state vector; Inputting the second information fusion vector into the self-attention submodule to obtain an initial second knowledge state vector; The third information fusion vector is input into the self-attention submodule to obtain an initial third knowledge state vector.

6. The method according to claim 4, wherein: Each sample exercise answer data in the sample exercise answer data sequence includes a sample exercise identifier; and determining a fourth loss value based on the sample exercise answer data sequence included in the training sample, the positive image exercise embedding matrix, the negative image exercise embedding matrix, the positive image user embedding matrix, and the negative image user embedding matrix, comprises: Determine a sample positive image user embedding vector and a sample negative image user embedding vector corresponding to the training sample based on the positive image user embedding matrix and the negative image user embedding matrix; For each sample exercise identifier included in the sample exercise answer data sequence in the training sample, the following steps are performed: Determine a sample positive image exercise embedding vector and a sample negative image exercise embedding vector corresponding to the sample exercise identifier based on the positive image exercise embedding matrix and the negative image exercise embedding matrix; Determine the dot product result of the sample positive image user embedding vector and the sample positive image exercise embedding vector as the correct answer prediction score; Determine the dot product result of the sample negative image user embedding vector and the sample negative image exercise embedding vector as the wrong answer prediction score; Determine the sample exercise identification, the predicted score of the correct answer and the predicted score of the incorrect answer as exercise score information; Based on the obtained score information of each exercise, a fourth loss value is generated.

7. The method according to claim 4, wherein: The method further comprises: In response to determining that the total loss value is not less than the preset threshold, relevant parameters in the initial knowledge tracking model are adjusted, and training samples are reselected from the training sample set, and the adjusted initial knowledge tracking model is used as the initial knowledge tracking model, and the training step is performed again.

8. A device for generating user exercise recommendation information, comprising: An acquisition unit is configured to acquire a user exercise interaction data sequence and an exercise concept association data sequence corresponding to each learning user, wherein each user exercise interaction data in the user exercise interaction data sequence includes an exercise answer data sequence; A generating unit configured to generate user exercise heterogeneous graph data, exercise concept hypergraph data and exercise directed transition graph data based on each exercise answer data sequence and the exercise concept association data sequence included in the user exercise interaction data sequence; An input unit is configured to input the user exercise interaction data sequence, the user exercise heterogeneous graph data, the exercise concept hypergraph data and the exercise directed transition graph data into a pre-trained knowledge tracking model to obtain each user knowledge state vector, wherein the user knowledge state vector in each user knowledge state vector corresponds to a learning user in each learning user, and the knowledge tracking model includes a sequence encoding module, a first graph embedding module, a second graph embedding module, a third graph embedding module, an interactive feature fusion module and a gated recurrent unit module; The execution unit is configured to execute the following steps for each of the learning users: Performing detection processing on each component in the user knowledge state vector corresponding to the learning user to obtain a first knowledge state component information sequence and a second knowledge state component information sequence; Generate a second knowledge state post-concept information set based on the second knowledge state component information sequence and the preset concept learning path information set; Based on the exercise concept association data sequence, the first knowledge state component information sequence and the second knowledge state post-concept information set, user exercise recommendation information is generated, and the user exercise recommendation information is stored in a cloud server for the learning user to obtain.

9. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images