Course teaching resource recommendation system and method based on big data

Abstract

The present application provides a course teaching resource recommendation system and method based on big data, which relates to the field of intelligent teaching. It uses data analysis and processing technology based on big data and deep learning to semantically embed and encode the acquired user course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically judge whether to push the teaching resources to be pushed according to the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, by understanding the contextual information of the user's browsing behavior in a more detailed and comprehensive manner, it is possible to provide teaching resources related to the user's current situation, thereby realizing a more intelligent course teaching resource recommendation.

Description

Technical Field

[0001] The present application relates to the field of intelligent teaching, and more specifically, to a course teaching resource recommendation system and method based on big data. Background Art

[0002] With the rapid development of online education and digital learning resources, users' demand for personalized learning experience is growing. Personalized recommendation systems can provide customized learning resources based on user behavior and preferences.

[0003] Chinese patent CN117575745B discloses a personalized recommendation method for course teaching resources based on AI big data, which first analyzes the behavior data of users browsing course resources, constructs a sample space and performs cluster analysis. Then, it calculates the user attention level of each cluster, identifies key features, extracts user preference keywords, and generates recommendation sequences based on these keywords to recommend teaching resources to users.

[0004] In the above-mentioned prior art, teaching resources in the teaching resource database are recommended to users in the order of the user's keyword recommendation sequence. Although this method avoids the "information cocoon" effect that may be caused by recommendation based solely on keywords, since the user's course resource browsing data contains rich semantic information and contextual connections, and this information is often more complex and richer than the content represented by a single keyword, if only relying on keywords, it is impossible to deeply understand the user's detailed needs and preferred contextual information, such as learning style, interactivity, etc., resulting in deviations in the recommendation results. In addition, users may need resources of different difficulty levels at different learning stages, and keyword-based recommendations may not be able to flexibly adapt to such changes.

[0005] Therefore, an optimized course teaching resource recommendation solution based on big data is needed. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a course teaching resource recommendation system and method based on big data, which uses data analysis and processing technology based on big data and deep learning to semantically embed and encode the acquired user course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically judge whether to push the teaching resources to be pushed according to the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, by understanding the contextual information of the user's browsing behavior in a more detailed and comprehensive way, it is possible to provide teaching resources related to the user's current situation, thereby realizing a more intelligent recommendation of course teaching resources.

[0007] According to one aspect of the present application, a course teaching resource recommendation system based on big data is provided, which includes:

[0008] User course resource browsing data acquisition module, used to obtain user course resource browsing data;

[0009] A user course resource browsing data semantic embedding coding module is used to perform semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of user course resource browsing record semantic embedding coding vectors;

[0010] A user course resource browsing record aggregation module is used to perform feature aggregation processing based on feature local significance gating on the set of semantic embedding coding vectors of the user course resource browsing record to obtain a user course resource browsing record aggregation representation vector;

[0011] A module for obtaining text descriptions of teaching resources to be pushed, used to obtain text descriptions of teaching resources to be pushed;

[0012] A semantic coding module for text description of teaching resources to be pushed, used for semantically coding the text description of the teaching resources to be pushed to obtain a semantic coding feature vector of the teaching resources to be pushed;

[0013] A module for semantic matching between the teaching resource to be pushed and the user course, used to calculate the semantic matching degree between the semantic encoding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record;

[0014] The teaching resource judgment module to be pushed is used to determine whether to push the teaching resource to be pushed based on the comparison between the semantic matching degree and a preset threshold.

[0015] According to another aspect of the present application, a method for recommending course teaching resources based on big data is provided, which includes:

[0016] Get user course resource browsing data;

[0017] Performing semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of semantic embedding coding vectors of the user course resource browsing records;

[0018] Performing feature aggregation processing based on feature local significance gating on the set of semantic embedding coding vectors of the user's course resource browsing records to obtain an aggregated representation vector of the user's course resource browsing records;

[0019] Get the text description of the teaching resources to be pushed;

[0020] Performing semantic coding on the text description of the teaching resource to be pushed to obtain a semantic coding feature vector of the teaching resource to be pushed;

[0021] Calculate the semantic matching degree between the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user's course resource browsing record;

[0022] Based on the comparison between the semantic matching degree and a preset threshold, it is determined whether to push the teaching resource to be pushed.

[0023] Compared with the prior art, the course teaching resource recommendation system and method based on big data provided by the present application uses data analysis and processing technology based on big data and deep learning to semantically embed and encode the acquired user course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically determine whether to push the teaching resources to be pushed according to the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, by understanding the contextual information of the user's browsing behavior in a more detailed and comprehensive manner, it is possible to provide teaching resources related to the user's current situation, thereby realizing a more intelligent course teaching resource recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0025] Figure 1 This is a system block diagram of a course teaching resource recommendation system based on big data according to an embodiment of the present application.

[0026] Figure 2 This is a data flow diagram of a course teaching resource recommendation system based on big data according to an embodiment of the present application.

[0027] Figure 3 This is a block diagram of a user course resource browsing record aggregation module in a course teaching resource recommendation system based on big data according to an embodiment of the present application.

[0028] Figure 4 This is a block diagram of a semantic feature energy level significance descriptor mask unit of a user course resource browsing record in a course teaching resource recommendation system based on big data according to an embodiment of the present application.

[0029] Figure 5This is a block diagram of a module for semantic matching of courses to be pushed to users in a course teaching resource recommendation system based on big data according to an embodiment of the present application.

[0030] Figure 6 The present invention is a flowchart of a method for recommending course teaching resources based on big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0032] With the rapid development of digital learning resources and online education, people's pursuit of personalized learning experience is constantly increasing. Personalized recommendation systems have emerged, which can provide tailored learning materials based on users' online behaviors and preferences.

[0033] Chinese patent CN117575745B discloses a personalized recommendation method for course teaching resources based on AI big data. It first creates a sample space by analyzing the behavioral data of users when browsing course resources, and applies cluster analysis methods. Then, it evaluates the degree of attention of users in each cluster, identifies key features, and extracts users' preferred keywords. Based on these keywords, the system generates a recommendation list and recommends corresponding teaching resources to users.

[0034] Although this method can avoid the "information cocoon" phenomenon that may be caused by relying solely on keyword recommendations, it still has limitations. Users' course browsing data contains rich semantic information and contextual connections, and the complexity of this information is far greater than that of a single keyword. Therefore, relying solely on keywords may not fully capture the contextual information of users' detailed needs and preferences, such as learning styles and interaction needs, which may lead to deviations in recommendation results. In addition, users may need resources of different difficulty levels at different stages of learning, and keyword-based recommendation systems may not be able to flexibly adapt to such changes in demand.

[0035] Therefore, in response to the above technical problems, the technical concept of the present application is to obtain the user's course resource browsing data and the text description of the teaching resources to be pushed, and use data analysis and processing technology based on big data and deep learning to semantically embed and encode the user's course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically determine whether to push the teaching resources to be pushed based on the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, the system can capture the user's subtle preferences more comprehensively, rather than just being limited to keywords. And the use of big data and deep learning can more carefully understand the contextual information of the user's browsing behavior, thereby providing teaching resource recommendations related to the user's current situation, so as to provide a more intelligent course teaching resource recommendation system.

[0036] Figure 1 This is a system block diagram of a course teaching resource recommendation system based on big data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of a course teaching resource recommendation system based on big data according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the course teaching resource recommendation system 100 based on big data, it includes: a user course resource browsing data acquisition module 110, which is used to acquire user course resource browsing data; a user course resource browsing data semantic embedding coding module 120, which is used to perform semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of user course resource browsing record semantic embedding coding vectors; a user course resource browsing record aggregation module 130, which is used to perform feature aggregation processing based on feature local significance gating on the set of user course resource browsing record semantic embedding coding vectors to obtain a user course resource browsing record aggregation representation vector. quantity; a module 140 for acquiring text descriptions of teaching resources to be pushed, for acquiring text descriptions of teaching resources to be pushed; a module 150 for semantically encoding text descriptions of teaching resources to be pushed, for semantically encoding the text descriptions of the teaching resources to be pushed to obtain semantically encoded feature vectors of the teaching resources to be pushed; a module 160 for semantically matching user courses to be pushed, for calculating the degree of semantic matching between the semantically encoded feature vectors of the teaching resources to be pushed and the aggregated representation vector of the user course resource browsing records; a module 170 for judging teaching resources to be pushed, for determining whether to push the teaching resources to be pushed based on a comparison between the degree of semantic matching and a preset threshold.

[0037] In an embodiment of the present application, the user course resource browsing data acquisition module 110 is used to obtain user course resource browsing data. It should be understood that the user course resource browsing data includes the user's browsing record data, specifically including the specific courses browsed by the user, the time the user stays on each course page, and the user's specific interactive behavior on the course page (such as clicking on course details, videos, exercises, etc.). In general, the user's browsing record data reflects the user's interests and preferences. By analyzing these data, it is possible to identify which types of courses the user prefers, and then push teaching resources that match these preferences, thereby achieving more intelligent course teaching resource recommendations. In particular, in a specific implementation method of an embodiment of the present application, the user's course resource browsing data can be acquired through the back-end database of the course teaching resource recommendation system.

[0038] In an embodiment of the present application, the user course resource browsing data semantic embedding coding module 120 is used to perform semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of semantic embedding coding vectors of the user course resource browsing records. Accordingly, in order to more deeply understand and grasp the key content and important points of interest browsed by the user, thereby providing more personalized resource recommendations, in the technical solution of the present application, semantic embedding coding is performed on each browsing record in the user course resource browsing data to capture and extract key semantic information, including the user's interest and needs in a specific topic or field, to obtain a set of semantic embedding coding vectors of the user course resource browsing records containing rich language representation information.

[0039] In an embodiment of the present application, the user course resource browsing record aggregation module 130 is used to perform feature aggregation processing based on feature local significance gating on the set of user course resource browsing record semantic embedding coding vectors to obtain the user course resource browsing record aggregation representation vector. Specifically, in an embodiment of the present application, the user course resource browsing record aggregation module is used to: input the set of user course resource browsing record semantic embedding coding vectors into a feature aggregation network based on feature energy local significance gating to obtain the user course resource browsing record aggregation representation vector. It should be understood that, considering that each user course resource browsing record semantic embedding coding feature in the set of user course resource browsing record semantic embedding coding vectors has different influences and semantic associations on the surrounding local semantic range. Therefore, in order to better extract and capture the user course resource browsing record semantic embedding coding feature information within each local semantic range, so as to capture the global features of user interests, in the technical solution of the present application, the set of user course resource browsing record semantic embedding coding vectors is input into a feature aggregation network based on feature energy local significance gating to obtain the user course resource browsing record aggregation representation vector.

[0040] It should be understood that the feature aggregation network based on local saliency gating of feature energy is a method that focuses on quantifying and aggregating local semantic energy distribution to refine feature representation, thereby improving the model's ability to comprehensively understand and analyze data. Specifically, first, based on the maximum, minimum, mean and variance of the semantic embedding encoding vector of each user's course resource browsing record, the feature energy level coefficient of each user's course resource browsing record semantic embedding encoding vector is calculated to quantify the energy level of the feature, and the obtained feature energy level coefficient can reflect the semantic importance and influence of the feature vector. Then, in order to capture the local saliency of the semantic features of the user's course resource browsing record, it is necessary to determine a reasonable neighborhood scale and calculate the local neighborhood saliency descriptor of each feature energy level coefficient to reflect the distribution of feature energy in the local neighborhood and reveal which features have more significant semantic information in this small range. Subsequently, the saliency descriptor is input into the mask module based on the gating function for adaptive dynamic adjustment of the feature saliency to obtain the mask-gated probabilistic user course resource browsing record semantic feature energy level saliency descriptor. In particular, the gating function (such as sigmoid) is used to perform nonlinear probabilistic transformation, while the masking function determines which features are retained and which are masked according to the transformed probability. Finally, the energy level significance descriptors of the semantic features of the user course resource browsing record of each masked gated probability are used as weights to perform weighted aggregation on the set of the user course resource browsing record semantic embedding coding vectors, which not only emphasizes the local significant user course resource browsing record semantic features, but also integrates the information of the semantic features of all user course resource browsing records to obtain the user course resource browsing record aggregate representation vector.

[0041] Specifically, Figure 3 FIG. 1 is a block diagram of a user course resource browsing record aggregation module in a course teaching resource recommendation system based on big data according to an embodiment of the present application. Figure 3As shown, the user course resource browsing record aggregation module 130 includes: a user course resource browsing record semantic feature energy level coefficient calculation unit 131, which is used to determine the feature energy level coefficient of each user course resource browsing record semantic embedding coding vector based on the maximum value, minimum value, mean value and variance of each user course resource browsing record semantic embedding coding vector in the set of user course resource browsing record semantic embedding coding vectors to obtain a set of user course resource browsing record semantic feature energy level coefficients; a user course resource browsing record semantic feature energy level significance descriptor calculation unit 132, which is used to determine the scale of the local neighborhood and calculate the local neighborhood of each user course resource browsing record semantic feature energy level coefficient in the set of user course resource browsing record semantic feature energy level coefficients. A saliency descriptor is used to obtain a set of semantic feature energy level saliency descriptors of user course resource browsing records; a user course resource browsing record semantic feature energy level saliency descriptor masking unit 133 is used to input the set of semantic feature energy level saliency descriptors of user course resource browsing records into a masking module based on a gating function to obtain a set of mask-gated probabilistic user course resource browsing record semantic feature energy level saliency descriptors; a user course resource browsing record aggregate representation generating unit 134 is used to use the set of mask-gated probabilistic user course resource browsing record semantic feature energy level saliency descriptors as a sequence of weights, and calculate the position-weighted sum of the set of semantic embedding coding vectors of the user course resource browsing records to obtain the user course resource browsing record aggregate representation vector.

[0042] More specifically, in an embodiment of the present application, the user course resource browsing record semantic feature energy level coefficient calculation unit is used to: respectively extract the maximum value and the minimum value in the semantic embedding coding vector of the user course resource browsing record to obtain the maximum value of the user course resource browsing record semantic feature and the minimum value of the user course resource browsing record semantic feature; respectively calculate the mean and the variance of the semantic embedding coding vector of the user course resource browsing record to obtain the mean value of the user course resource browsing record semantic feature and the variance of the user course resource browsing record semantic feature; add the variance of the user course resource browsing record semantic feature and the preset hyperparameter to obtain the first energy level coefficient of the user course resource browsing record semantic feature; calculate the maximum value of the user course resource browsing record semantic feature and the variance of the user course resource browsing record semantic feature. The square of the difference between the mean values ​​of the semantic features of the user's course resource browsing records is calculated to obtain the maximum semantic difference value of the user's course resource browsing records; the square of the difference between the minimum value of the semantic features of the user's course resource browsing records and the mean value of the semantic features of the user's course resource browsing records is calculated to obtain the minimum semantic difference value of the user's course resource browsing records; the maximum semantic difference value of the user's course resource browsing records is added to the minimum semantic difference value of the user's course resource browsing records and the preset hyperparameter to obtain the second energy level coefficient of the semantic features of the user's course resource browsing records; the first energy level coefficient of the semantic features of the user's course resource browsing records is divided by the second energy level coefficient of the semantic features of the user's course resource browsing records to obtain the energy level coefficient of the semantic features of the user's course resource browsing records.

[0043] More specifically, Figure 4 The block diagram of the user course resource browsing record semantic feature energy level significance descriptor mask unit in the course teaching resource recommendation system based on big data according to the embodiment of the present application. Figure 4 As shown, the user course resource browsing record semantic feature energy level significance descriptor mask unit 133 includes: a browsing record semantic feature energy level significance descriptor normalization processing subunit 1331, which is used to normalize each user course resource browsing record semantic feature energy level significance descriptor in the set of user course resource browsing record semantic feature energy level significance descriptors to obtain a set of normalized user course resource browsing record semantic feature energy level significance descriptors; a browsing record semantic feature energy level significance descriptor masking probabilization subunit 1332, which is used to input each normalized user course resource browsing record semantic feature energy level significance descriptor in the set of normalized user course resource browsing record semantic feature energy level significance descriptors into a gating function for masking processing to obtain the set of mask-gated probabilistic user course resource browsing record semantic feature energy level significance descriptors.

[0044] More specifically, in an embodiment of the present application, the browsing record semantic feature energy level significance descriptor mask probabilization subunit is used to: set each normalized user course resource browsing record semantic feature energy level significance descriptor greater than a predetermined threshold in the set of normalized user course resource browsing record semantic feature energy level significance descriptors to an original value, and set the rest to zero to obtain the set of masked gated probabilistic user course resource browsing record semantic feature energy level significance descriptors.

[0045] In the embodiment of the present application, specifically, the user course resource browsing record aggregation module is used to: input the set of semantic embedding coding vectors of the user course resource browsing record into a feature aggregation network based on feature energy local significance gating, and process it with the following feature aggregation formula to obtain the user course resource flow record aggregation representation vector; wherein, the feature aggregation formula is:

[0046] X={x1,x2,...,x k ,...,x n}

[0047]

[0048]

[0049] w si =mask(w i )

[0050]

[0051] Where X is the set of semantic embedding coding vectors of the user’s course resource browsing records, x k is the semantic embedding encoding vector of the kth user’s course resource browsing record, x n is the semantic embedding coding vector of the nth user course resource browsing record, n represents the number of feature vectors in the set of semantic embedding coding vectors of the user course resource browsing record, μ i and σ 2 i are the mean and variance of the semantic embedding encoding vector of the i-th user's course resource browsing record, max(x i ) and min(x i ) are the maximum and minimum values ​​in the semantic embedding encoding vector of the i-th user’s course resource browsing record, ε is the preset hyperparameter, e i is the characteristic energy level coefficient of the semantic embedding encoding vector of the i-th user’s course resource browsing record, e q is the characteristic energy level coefficient of the semantic embedding encoding vector of the qth user course resource browsing record, Num rIndicates that the i The number of characteristic energy coefficients in the local neighborhood centered at ci is the semantic feature energy level significant descriptor of the i-th user course resource browsing record in the set of semantic feature energy level significant descriptors of the user course resource browsing record, w i is the ith normalized user course resource browsing record semantic feature energy level significant descriptor in the set of normalized user course resource browsing record semantic feature energy level significant descriptors, mask(·) is masking processing, w si is the ith mask-gated probabilistic user course resource browsing record semantic feature energy level significance descriptor in the set of mask-gated probabilistic user course resource browsing record semantic feature energy level significance descriptors, x i is the semantic embedding encoding vector of the i-th user’s course resource browsing record, θ is the predetermined threshold, and v c An aggregate representation vector is created for the course resource flow records of the user.

[0052] In an embodiment of the present application, the module 140 for acquiring the text description of the teaching resources to be pushed is used to acquire the text description of the teaching resources to be pushed. It should be understood that the text description of the teaching resources to be pushed contains information related to the teaching resources such as the specific name of the course, the type of course, and an overview of the course content. By acquiring a detailed text description of the teaching resources to be pushed and performing a semantic coding analysis on it, the system can better understand the content and characteristics of the course, thereby making more accurate judgments when pushing. This content-based recommendation method not only improves the relevance of the recommendation, but also can better meet the learning needs of users. In particular, in a specific implementation method of the embodiment of the present application, the acquisition of the text description data of the teaching resources to be pushed can be performed through the back-end database of the course teaching resource recommendation system.

[0053] In an embodiment of the present application, the semantic encoding module 150 of the text description of the teaching resources to be pushed is used to semantically encode the text description of the teaching resources to be pushed to obtain the semantic encoding feature vector of the teaching resources to be pushed. Specifically, in an embodiment of the present application, the semantic encoding module of the text description of the teaching resources to be pushed is used to: after the text description of the teaching resources to be pushed is processed by word segmentation, it is input into the semantic encoder of the teaching resources to be pushed containing the word embedding layer to obtain the semantic encoding feature vector of the teaching resources to be pushed. It should be understood that in order to understand and more accurately analyze the semantic feature information of the text description of the teaching resources to be pushed, in the technical solution of the present application, the text description of the teaching resources to be pushed is semantically encoded to obtain the semantic encoding feature vector of the teaching resources to be pushed. That is, semantic encoding can learn the key semantic correlations between words in the text description of the teaching resources to be pushed, so as to better express the meaning of the text, thereby improving the relevance and personalization of the teaching resource recommendation.

[0054] In an embodiment of the present application, the to-be-pushed-user course semantic matching module 160 is used to calculate the semantic matching degree between the to-be-pushed teaching resource semantic encoding feature vector and the user course resource browsing record aggregate representation vector. It should be understood that the user course resource browsing record aggregate representation vector is constructed based on the user's historical browsing behavior and reflects the user's interest in different types of content. The to-be-pushed teaching resource semantic encoding feature vector is extracted from the description of the teaching resource, which captures the theme, difficulty level and other relevant information of the resource. Therefore, in order to help the system to more accurately understand the user's current interests and recommend the most relevant teaching resources accordingly, in the technical solution of the present application, the semantic matching degree between the to-be-pushed teaching resource semantic encoding feature vector and the user course resource browsing record aggregate representation vector is calculated. In particular, the semantic matching degree is first obtained by calculating the difference vector between the to-be-pushed teaching resource semantic encoding feature vector and the user course resource browsing record aggregate representation vector to quantify the semantics between the two quantities, and at the same time calculating the covariance matrix between the to-be-pushed teaching resource semantic encoding feature vector and the user course resource browsing record aggregate representation vector, and obtaining the semantic matching degree between the two based on the difference feature and the covariance matrix.

[0055] Specifically, Figure 5 FIG. 1 is a block diagram of a semantic matching module for courses to be pushed and for user courses in a course teaching resource recommendation system based on big data according to an embodiment of the present application. Figure 5As shown, the to-be-pushed-user course semantic matching module 160 includes: a to-be-pushed-user course resource differential feature calculation unit 161, which is used to calculate the positional difference between the to-be-pushed teaching resource semantic coding feature vector and the user course resource browsing record aggregate representation vector to obtain the to-be-pushed-user course resource differential feature vector; a to-be-pushed-user course resource inverse covariance matrix generation unit 162, which is used to calculate the covariance matrix between the to-be-pushed teaching resource semantic coding feature vector and the user course resource browsing record aggregate representation vector, and calculate the inverse of the covariance matrix to obtain the to-be-pushed-user course resource inverse covariance matrix; a to-be-pushed-user course resource matching related value calculation unit 163, which is used to calculate the transposed vector of the to-be-pushed-user course resource differential feature vector and the product of the to-be-pushed-user course resource inverse covariance matrix and the to-be-pushed-user course resource differential feature vector to obtain the to-be-pushed-user course resource matching related value; a semantic matching degree result generation unit 164, which is used to calculate the square root of the to-be-pushed-user course resource matching related value to obtain the semantic matching degree.

[0056] In the embodiment of the present application, specifically, the to-be-pushed-user course semantic matching module is used to calculate the semantic matching degree between the semantic encoding feature vector of the to-be-pushed teaching resource and the aggregate representation vector of the user course resource browsing record using the following semantic matching formula; wherein the semantic matching formula is:

[0057]

[0058] Among them, x i is the semantic encoding feature vector of the teaching resource to be pushed, v c Aggregate representation vector for the user's course resource browsing records, (·) T is the transpose of the vector, S is the covariance matrix between the semantic encoding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user's course resource browsing record, D(x i ,v c ) is the semantic matching degree.

[0059] In an embodiment of the present application, the teaching resource judgment module 170 to be pushed is used to determine whether to push the teaching resource to be pushed based on the comparison between the semantic matching degree and the preset threshold. That is, the semantic matching coefficient calculated between the semantic encoding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user's course resource browsing record is compared with the preset threshold to automatically determine whether to push the teaching resource to be pushed. In this way, the system can capture the user's subtle preferences more comprehensively, rather than just being limited to keywords. And the use of big data and deep learning can more carefully understand the contextual information of the user's browsing behavior, thereby providing teaching resource recommendations related to the user's current situation, so as to provide a more intelligent course teaching resource recommendation system.

[0060] In particular, in a preferred embodiment, the present application takes into account that the semantic coding feature vector of the teaching resources to be pushed and the aggregated representation vector of the user's course resource browsing records respectively represent the text semantic coding features of the text description of the teaching resources to be pushed and the aggregated text semantic features of each browsing record in the user's course resource browsing data based on the significance of the local sample domain semantic energy distribution. Therefore, when the semantic matching degree between them is calculated and mapped to the common semantic matching domain, the semantic coding feature vector of the teaching resources to be pushed and the aggregated representation vector of the user's course resource browsing records will also cause matching semantic mapping offset due to the difference in semantic feature expression dimensions, resulting in insufficient coverage of the offset semantic mapping domain, causing outlier regression inference mapping deviation, and affecting the accuracy of the semantic matching degree between the semantic coding feature vector of the teaching resources to be pushed and the aggregated representation vector of the user's course resource browsing records.

[0061] Based on this, when calculating the semantic matching degree between the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record, the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record are optimized. The optimization process specifically includes: cascading the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record into a semantic-user course resource browsing record aggregate feature vector of the teaching resource to be pushed; calculating the feature mean of the semantic-user course resource browsing record aggregate feature vector of the teaching resource to be pushed, and dividing the feature mean by the teaching resource to be pushed. The difference between the maximum eigenvalue and the minimum eigenvalue of the aggregated feature vector of resource semantics-user course resource browsing records is used to obtain the aggregated distribution representation value of the teaching resource semantics-user course resource browsing records to be pushed; the value of the aggregated distribution representation of the teaching resource semantics-user course resource browsing records to be pushed is subtracted from one and then divided by the aggregated distribution representation value of the teaching resource semantics-user course resource browsing records to be pushed to obtain the aggregated distribution modulation value of the teaching resource semantics-user course resource browsing records to be pushed; the aggregated feature vector of the teaching resource semantics-user course resource browsing records to be pushed is activated by a probabilistic function to obtain the probabilistic teaching resource semantics-user course resource to be pushed Browsing record aggregation feature vector; after point subtraction of the probabilistic teaching resource semantics to be pushed - user course resource browsing record aggregation feature vector and the teaching resource semantics to be pushed - user course resource browsing record aggregation distribution modulation value, take the absolute value and calculate the negative of the logarithmic value with base 2 to obtain the probabilistic teaching resource semantics to be pushed - user course resource browsing record aggregation distribution modulation information vector; divide the teaching resource semantics to be pushed - user course resource browsing record aggregation distribution representation value by one minus the difference of each feature value of the probabilistic teaching resource semantics to be pushed - user course resource browsing record aggregation feature vector, for the probabilistic teaching resource semantics to be pushed - user course resource browsing record aggregation distribution modulation information vector Sum all the eigenvalues ​​of the teaching resource semantics-user course resource browsing record aggregation feature vector, and divide it by the length of the teaching resource semantics-user course resource browsing record aggregation feature vector to be pushed to obtain the probabilistic teaching resource semantics-user course resource browsing record aggregation distribution modulation bias value to be pushed; perform dot-addition on the probabilistic teaching resource semantics-user course resource browsing record aggregation distribution modulation information vector to be pushed and the probabilistic teaching resource semantics-user course resource browsing record aggregation distribution modulation bias value to be pushed and the weight as a hyperparameter to obtain the optimized teaching resource semantics-user course resource browsing record aggregation feature vector to be pushed;The optimized teaching resource semantics to be pushed-user course resource browsing record aggregate feature vector is split into the optimized teaching resource semantics to be pushed feature vector and the optimized user course resource browsing record aggregate representation vector according to the cascade of the teaching resource semantics encoding feature vector to be pushed and the user course resource browsing record aggregate representation vector. ;

[0062] Among them, the optimized representation of the aggregated feature vector of the teaching resource semantics to be pushed and the user course resource browsing record is:

[0063]

[0064] in, is the feature mean of the aggregated feature vector of the semantics of the teaching resource to be pushed and the user course resource browsing record, v max and v min are the maximum eigenvalue and the minimum eigenvalue of the aggregated feature vector of the semantics of the teaching resources to be pushed-user course resource browsing records, p is the aggregated distribution representation value of the semantics of the teaching resources to be pushed-user course resource browsing records, V represents the probabilistic aggregated feature vector of the semantics of the teaching resources to be pushed-user course resource browsing records, Indicates point-by-point subtraction, log indicates the logarithmic function value with base 2, v i represents the feature value of the ith position of the probabilistic teaching resource semantics to be pushed-user course resource browsing record aggregation feature vector, It represents adding by position point, α is the weight as a hyperparameter, L is the length of the aggregated feature vector of the semantics of the teaching resources to be pushed-the user's course resource browsing record, and V' represents the optimized aggregated feature vector of the semantics of the teaching resources to be pushed-the user's course resource browsing record.

[0065] That is, in the above preferred example, the probability information distribution planning based on the eigenvalue of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record is carried out through the Bernoulli probability modulation distribution of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record relative to the eigenvalue distribution, and the probability reverse mapping of the probability characteristics of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record is used as the extended coverage of the set mapping space of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record, so as to independently understand the intuitive probability information distribution and abstract probability space mapping of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record, so as to improve the calculation accuracy of the semantic matching degree between the optimized semantic coding feature vector of the teaching resource to be pushed and the optimized aggregate representation vector of the user course resource browsing record by avoiding the counterfactual reasoning mapping of the outlier feature distribution of the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record to the regression probability. In this way, the system can capture the user's subtle preferences more comprehensively, rather than just being limited to keywords. And the use of big data and deep learning can more carefully understand the contextual information of the user's browsing behavior, thereby providing teaching resource recommendations related to the user's current situation, and providing a more intelligent course teaching resource recommendation system.

[0066] In summary, the course teaching resource recommendation system 100 based on big data according to the embodiment of the present application is explained, which uses data analysis and processing technology based on big data and deep learning to semantically embed and encode the acquired user course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically judge whether to push the teaching resources to be pushed according to the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, by understanding the contextual information of the user's browsing behavior in a more detailed and comprehensive manner, it is possible to provide teaching resources related to the user's current situation, thereby realizing a more intelligent recommendation of course teaching resources.

[0067] As described above, the course teaching resource recommendation system 100 based on big data according to the embodiment of the present application can be implemented in various terminal devices, such as a server for recommending course teaching resources based on big data. In one example, the course teaching resource recommendation system 100 based on big data according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the course teaching resource recommendation system 100 based on big data can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the course teaching resource recommendation system 100 based on big data can also be one of the many hardware modules of the terminal device.

[0068] Alternatively, in another example, the course teaching resource recommendation system 100 based on big data and the terminal device may also be separate devices, and the course teaching resource recommendation system 100 based on big data may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0069] Figure 6 Flow chart of a method for recommending course teaching resources based on big data according to an embodiment of the present application. Figure 6 As shown, in the course teaching resource recommendation method based on big data, it includes: S110, obtaining user course resource browsing data; S120, semantically embedding coding each browsing record in the user course resource browsing data to obtain a set of user course resource browsing record semantic embedding coding vectors; S130, performing feature aggregation processing based on feature local significance gating on the set of user course resource browsing record semantic embedding coding vectors to obtain a user course resource browsing record aggregate representation vector; S140, obtaining a text description of the teaching resource to be pushed; S150, semantically encoding the text description of the teaching resource to be pushed to obtain a semantic encoding feature vector of the teaching resource to be pushed; S160, calculating the semantic matching degree between the semantic encoding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record; S170, determining whether to push the teaching resource to be pushed based on the comparison between the semantic matching degree and a preset threshold.

[0070] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned course teaching resource recommendation method based on big data have been referred to above. Figures 1 to 5 It has been introduced in detail in the description of the course teaching resource recommendation system based on big data, and therefore, its repeated description will be omitted.

[0071] In summary, a course teaching resource recommendation method based on big data according to an embodiment of the present application is explained, which uses data analysis and processing technology based on big data and deep learning to semantically embed and encode the acquired user course resource browsing data and the text description of the teaching resources to be pushed, so as to automatically judge whether to push the teaching resources to be pushed according to the language matching degree between the aggregated representation features of each user's course resource browsing record in the entire browsing process and the semantic coding features of the teaching resources to be pushed and the comparison with the preset threshold. In this way, by understanding the contextual information of the user's browsing behavior in a more detailed and comprehensive manner, it is possible to provide teaching resources related to the user's current situation, thereby achieving a more intelligent course teaching resource recommendation.

[0072] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of the technical solution of the present application.

Claims

1. A course teaching resource recommendation system based on big data, characterized in that: include: User course resource browsing data acquisition module, used to obtain user course resource browsing data; A user course resource browsing data semantic embedding coding module is used to perform semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of user course resource browsing record semantic embedding coding vectors; A user course resource browsing record aggregation module is used to perform feature aggregation processing based on feature local significance gating on the set of semantic embedding coding vectors of the user course resource browsing record to obtain a user course resource browsing record aggregation representation vector; A module for obtaining text descriptions of teaching resources to be pushed, used to obtain text descriptions of teaching resources to be pushed; A semantic coding module for text description of teaching resources to be pushed, used for semantically coding the text description of the teaching resources to be pushed to obtain a semantic coding feature vector of the teaching resources to be pushed; A module for semantic matching between the teaching resource to be pushed and the user course, used to calculate the semantic matching degree between the semantic encoding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user course resource browsing record; A teaching resource judgment module to be pushed, used for determining whether to push the teaching resource to be pushed based on the comparison between the semantic matching degree and a preset threshold; The user course resource browsing record aggregation module is used to: input the set of semantic embedding coding vectors of the user course resource browsing record into a feature aggregation network based on feature energy local significance gating to obtain the user course resource browsing record aggregation representation vector; The user course resource browsing record aggregation module includes: A user course resource browsing record semantic feature energy level coefficient calculation unit, used to determine the feature energy level coefficient of each user course resource browsing record semantic embedding coding vector in the set of user course resource browsing record semantic embedding coding vectors based on the maximum value, minimum value, mean value and variance of each user course resource browsing record semantic embedding coding vector to obtain a set of user course resource browsing record semantic feature energy level coefficients; A user course resource browsing record semantic feature energy level significance descriptor calculation unit, used to determine the scale of the local neighborhood, calculate the local neighborhood significance descriptor of each user course resource browsing record semantic feature energy level coefficient in the set of user course resource browsing record semantic feature energy level coefficients to obtain a set of user course resource browsing record semantic feature energy level significance descriptors; A user course resource browsing record semantic feature energy level significance descriptor mask unit, used to input the set of user course resource browsing record semantic feature energy level significance descriptors into a mask module based on a gating function to obtain a mask-gated probabilistic user course resource browsing record semantic feature energy level significance descriptor set; A user course resource browsing record aggregate representation generating unit, used to calculate the position-weighted sum of the set of semantic embedding coding vectors of the user course resource browsing record using the set of the mask-gated probabilistic user course resource browsing record semantic feature energy level significance descriptors as a sequence of weights to obtain the user course resource browsing record aggregate representation vector; Wherein, the user course resource browsing record semantic feature energy level coefficient calculation unit is used to: Respectively extract the maximum value and the minimum value in the semantic embedding coding vector of the user course resource browsing record to obtain the maximum value of the semantic feature of the user course resource browsing record and the minimum value of the semantic feature of the user course resource browsing record; Calculate the mean and variance of the semantic embedding coding vector of the user's course resource browsing record respectively to obtain the mean of the semantic features of the user's course resource browsing record and the variance of the semantic features of the user's course resource browsing record; Adding the semantic feature variance of the user's course resource browsing record and the preset hyperparameter to obtain the first energy level coefficient of the semantic feature of the user's course resource browsing record; Calculate the square of the difference between the maximum value of the semantic feature of the user's course resource browsing record and the mean value of the semantic feature of the user's course resource browsing record to obtain the maximum semantic difference value of the user's course resource browsing record; Calculate the square of the difference between the minimum value of the semantic feature of the user's course resource browsing record and the mean value of the semantic feature of the user's course resource browsing record to obtain the minimum semantic difference value of the user's course resource browsing record; The maximum semantic difference value of the user's course resource browsing record is added to the minimum semantic difference value of the user's course resource browsing record and the preset hyperparameter to obtain the second energy level coefficient of the semantic feature of the user's course resource browsing record; The first energy level coefficient of the semantic feature of the user's course resource browsing record is divided by the second energy level coefficient of the semantic feature of the user's course resource browsing record to obtain the energy level coefficient of the semantic feature of the user's course resource browsing record.

2. According to the course teaching resource recommendation system based on big data according to claim 1, it is characterized in that: The user course resource browsing record semantic feature energy level significance descriptor mask unit includes: The browsing record semantic feature energy level significance descriptor normalization processing subunit is used to normalize each user course resource browsing record semantic feature energy level significance descriptor in the set of user course resource browsing record semantic feature energy level significance descriptors to obtain a set of normalized user course resource browsing record semantic feature energy level significance descriptors; The browsing record semantic feature energy level significance descriptor mask probabilization subunit is used to input each normalized user course resource browsing record semantic feature energy level significance descriptor in the set of normalized user course resource browsing record semantic feature energy level significance descriptors into a gating function for masking processing to obtain the set of masked gated probabilistic user course resource browsing record semantic feature energy level significance descriptors.

3. The course teaching resource recommendation system based on big data according to claim 2 is characterized in that: The browsing record semantic feature energy level significance descriptor mask probabilization subunit is used to: set each normalized user course resource browsing record semantic feature energy level significance descriptor greater than a predetermined threshold in the set of normalized user course resource browsing record semantic feature energy level significance descriptors to the original value, and set the rest to zero to obtain the set of masked gated probabilistic user course resource browsing record semantic feature energy level significance descriptors.

4. The course teaching resource recommendation system based on big data according to claim 3 is characterized in that: The semantic encoding module for the text description of the teaching resources to be pushed is used to: perform word segmentation processing on the text description of the teaching resources to be pushed and input it into the semantic encoder of the teaching resources to be pushed containing a word embedding layer to obtain the semantic encoding feature vector of the teaching resources to be pushed.

5. The course teaching resource recommendation system based on big data according to claim 4 is characterized in that: The module for semantic matching of user courses to be pushed includes: A differential feature calculation unit for user course resources to be pushed, used for calculating the positional difference between the semantic coding feature vector of the teaching resources to be pushed and the aggregate representation vector of the user course resource browsing records to obtain a differential feature vector for user course resources to be pushed; The inverse covariance matrix generating unit of the user course resources to be pushed is used to calculate the covariance matrix between the semantic coding feature vector of the teaching resources to be pushed and the aggregate representation vector of the user course resource browsing record, and calculate the inverse of the covariance matrix to obtain the inverse covariance matrix of the user course resources to be pushed; A unit for calculating the matching correlation value between the to-be-pushed and user course resources, used for calculating the transposed vector of the to-be-pushed and user course resource differential feature vector and the product of the inverse covariance matrix of the to-be-pushed and user course resource differential feature vector to obtain the matching correlation value between the to-be-pushed and user course resources; The semantic matching result generating unit is used to calculate the square root of the matching correlation value of the resource to be pushed and the user course resource to obtain the semantic matching degree.

6. A method for recommending course teaching resources based on big data, using the course teaching resource recommendation system based on big data according to claim 1, characterized in that: include: Get user course resource browsing data; Performing semantic embedding coding on each browsing record in the user course resource browsing data to obtain a set of semantic embedding coding vectors of the user course resource browsing records; Performing feature aggregation processing based on feature local significance gating on the set of semantic embedding coding vectors of the user's course resource browsing records to obtain an aggregated representation vector of the user's course resource browsing records; Get the text description of the teaching resources to be pushed; Performing semantic coding on the text description of the teaching resource to be pushed to obtain a semantic coding feature vector of the teaching resource to be pushed; Calculate the semantic matching degree between the semantic coding feature vector of the teaching resource to be pushed and the aggregate representation vector of the user's course resource browsing record; Based on the comparison between the semantic matching degree and a preset threshold, it is determined whether to push the teaching resource to be pushed.

Images