Diversified course recommendation method and system based on learning course interval time

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a diversified course recommendation method and system based on learning course interval time. The method comprises the steps of S1, constructing a course feature model based on a time interval of a user interaction course, S2, constructing a course diversification recommendation model based on a capsule network technology, S3, carrying out model training and model testing, and S4, using the model. Technical methods such as a self-attention mechanism and a dynamic routing mechanism are utilized, historical learning record data of students are systematically analyzed, time intervals and sequence position features between courses in interaction are fused into student learning sequence modeling, meanwhile, diversified learning interests and learning requirements of learners are considered, and the learning efficiency is improved. And the accuracy and interpretability of course recommendation are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a diversified course recommendation method and system based on the interval time of learning courses. Background Art

[0002] With the rapid development of online education, more and more students begin to use various online learning platforms to freely select courses they are interested in and formulate personalized learning plans. However, the massive course resources bring the problem of information overload, making it difficult for students to find suitable courses for themselves. To solve this problem, course recommendation algorithms have become one of the important research directions, and their goal is to analyze students' historical learning behaviors and interaction data, mine their interest preferences, provide targeted learning resources for students, and improve their learning efficiency.

[0003] Existing course recommendation methods mainly include models based on collaborative filtering, models based on graph neural networks, and models based on reinforcement learning, etc. These methods provide personalized recommendation results by combining students' historical behaviors, course contents, and student characteristics. Although these course recommendation methods have achieved good results, they still face certain problems.

[0004] Most existing methods mainly focus on studying the sequential relationship of courses in the learner interaction sequence, while ignoring the key feature of time interval. However, students' preferences change dynamically with time and historical interaction courses, and the time interval feature between courses in the interaction reflects students' learning behavior patterns and interest evolution trends. Existing recommendation models usually generate a single overall embedding based on the student's interaction sequence to represent the student. These models usually assume that all courses in an interaction sequence correspond to a main interest. However, in the real world, learners' needs and behaviors are often diverse, and their interaction behaviors usually involve multiple different aspects of interests, corresponding to different types of courses. Modeling a single interest cannot accurately depict learners' diverse preferences. Summary of the Invention

[0005] (I) Technical Problems to be Solved The main purpose of the present invention is to provide a diversified course recommendation method and system based on the interval time of learning courses to solve the above problems.

[0006] (II) Technical Solutions To achieve the above object, a diversified course recommendation method based on the interval time of learning courses provided by the present invention includes the steps of: S1, constructing a course feature model based on the time interval of user interaction courses, including: S11, feature embedding representation: According to the course interaction length sequence of learner u and the course interaction time series , construct a course embedding matrix to represent all courses in the interaction sequence, and construct a relative course time interval embedding matrix to represent the time interval information between courses, and construct an embedding matrix of course positions to represent the sequential information in the course sequence; where represents the nth element in the course interaction length sequence, represents the nth element in the course interaction time series; S12, construct a time interval perception module: Based on the self-attention mechanism, according to the course embedding matrix and the relative course time interval embedding matrix construct a time interval perception module to obtain a representation with relative course time interval features ; where is the hidden dimension, is an attention function set to prevent and from having too high a weight score for the inner product, , , , , is a trainable weight parameter matrix; S13, construct a sequential position perception module: Based on the self-attention mechanism, according to the course position embedding matrix construct a sequential position perception module to obtain a representation with sequential position features ; where d is the hidden dimension, h is the attention head set to prevent and from having too high a weight score for the inner product, , is a trainable weight parameter matrix; S14, feature fusion representation: Linearly weight and fuse the course representation with relative time interval features and the representation with sequential position features to obtain a fused course feature , and then perform a non-linear transformation to obtain the final course feature representation ; where ; S2, construct a course diversification recommendation model based on capsule network technology, including: S21, construct a two-layer capsule network, including low-level capsule i and high-level capsule j: The embedding of the vector of the course feature representation ​The low-level capsule i is used to represent the course sequence features; the high-level capsule j is used to represent different interest course types; where , the low-level capsule and the high-level capsule The routing logic is as follows: , where is a bilinear mapping matrix. The high-level capsule is used to capture the relationship between course embeddings and interests; S22, determine the total input vector candidate vector of the high-level capsule through dynamic routing ; S23, output the high-level capsule: for the total input vector candidate vector , through the formula: , generate the high-level capsule output v j ; S24, for the learner , the high-level capsule output forms a matrix ; S3, perform model training and model testing on the models of step S1 and step S2 to train and adjust the parameters, and obtain the finally determined models of step S1 and step S2; S4, use the models of step S1 and step S2 finally determined by step S3 to make diversified course recommendations for users.

[0007] Preferably, in step S11, it further includes: when the length of the course interaction length sequence exceeds the preset length n, only retain the latest n courses in the course interaction length sequence; when the length of the course interaction length sequence is less than the preset length n, fill the front of the course interaction length sequence with specific items until the length of the course interaction length sequence reaches n; where n is a positive integer, and the specific item is a value with n being 1; When the length of the course interaction time sequence exceeds the preset length n, only retain the latest n courses in the course interaction time sequence; when the length of the course interaction time sequence is less than the preset length n, fill the front of the course interaction time sequence with specific items until the length of the course interaction time sequence reaches n; where n is a positive integer, and the specific item is a value with n being 1.

[0008] Preferably, the course embedding matrix , where is the dimension of the embedding vector, represents the embedding of the -th course in the sequence; The relative course time interval embedding matrix , where , represents the relative time interval between course and course ; The embedding matrix of the course position , where is a learnable position embedding matrix, represents the embedding of the -th course in the sequence.

[0009] Preferably, the non-linear transformation in step S14 includes: S141, inputting the fused course features into a feed-forward neural network combined with the ReLU activation function: where , is the corresponding weight matrix, , is the corresponding bias vector; S142, further optimizing through layer normalization, residual connection, and dropout regularization techniques to obtain the final feature representation : where is the element-wise product, and are the mean and variance used to standardize the variable, and are the learned scale factor and bias term, is the i-th element of the standardized variable, is a set small constant.

[0010] Preferably, the total input vector candidate vector , where is the coupling coefficient determined by the iterative dynamic routing process; , where represents the logarithmic prior probability that capsule should be coupled with capsule .

[0011] Preferably, the model training in step S3 includes: S31. Extract the interest vector of the user that best matches the target course , , where is the multi - interest representation matrix of the user, represents the embedding vector of the target course; S32. Set the training sample that includes the user interest representation and the target course embedding , and calculate the probability of interaction between the user and the target course through inner product: : S33. Calculate the loss function for model training: S34. Optimize the model parameters in steps S1 and S2 based on the loss function.

[0012] Preferably, the model testing in step S3 includes: Measuring the quality of the recommendation by calculating the overall interest degree of the user in the recommended courses , where the is: where represents the total interest degree of the user in the recommendation set , is the embedding vector of the candidate course, represents the th interest vector of the learner

[0013] Preferably, step S4 includes deploying the models in steps S1 and S2 finally determined through step S3 to the actual application scenario to achieve diversified course recommendation based on the learning course interval time.

[0014] The present invention also provides a diversified course recommendation system based on the learning course interval time, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the diversified course recommendation method as described in any one of the above are implemented.

[0015] The present invention also provides a computer - readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the diversified course recommendation method as described in any one of the above are implemented.​

[0016] (3) Beneficial effects A diversified course recommendation method based on the time interval of learning courses proposed in this application takes the time interval of learners' course learning as a consideration factor for course recommendation prediction, more accurately understands students' learning habits and needs, and the detailed analysis enables the system to recommend courses that are more in line with their current learning status and interests for students, avoiding the blindness and randomness that may exist in traditional recommendation methods. At the same time, considering the diversified learning interests and learning needs of learners, a multi-interest extraction layer is designed based on a dynamic routing mechanism, which adaptively extracts multiple interest vectors from the historical interaction sequence of learners, and accurately depicts the diversified interests of learners in different learning scenarios by dynamically clustering the characteristics of different interaction behaviors, further improving the accuracy and interpretability of course recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, as part of the present invention, are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute an improper limitation to the present invention. Obviously, the accompanying drawings in the following description are only some embodiments, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. In the drawings: Figure 1 Schematic diagram of a feature embedding representation mechanism provided in this embodiment; Figure 2 Schematic diagram of a time interval perception module model and a sequential position perception module model provided in this embodiment; Figure 3 Schematic diagram of a diversified course recommendation model based on capsule network technology provided in this embodiment; Figure 4 Schematic diagram of a model training and model testing process provided in this embodiment; Figure 5 Schematic diagram of a framework of a diversified course recommendation system based on the time interval of learning courses provided in this embodiment; Figure 6 Schematic diagram of a process of a diversified course recommendation method based on the time interval of learning courses provided in this embodiment; Figure 7 Schematic diagram of a hardware structure for running a diversified course recommendation system based on the time interval of learning courses provided in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the accompanying drawings through specific embodiments.

[0019] It should be noted that all the directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement conditions between components in a certain specific posture (as shown in the drawings). If the specific posture changes, the directional indications will also change accordingly.

[0020] In addition, in the present invention, descriptions such as "first" and "second" are only for descriptive purposes and should not be construed as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically defined.

[0021] In the present invention, unless otherwise clearly specified and limited, terms such as "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or integrated; "connection" can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and can be the communication inside two components or the interaction relationship between two components, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0022] As Figure 1 - Figure 6 shown, in this embodiment, a diversified course recommendation method based on the interval time of learning courses is provided, including steps S1 - S4.

[0023] S1. Construct a course feature model based on the time interval of user interaction courses. Design a time interval perception module based on the self - attention mechanism, which is used to dynamically model the time interval characteristics in the learning behavior of learners. At the same time, introduce a position perception module to model the sequential position information in the interaction sequence. By fusing the time interval and sequential position information, a comprehensively weighted sequence representation is generated. The model with the time interval of user interaction courses as an evaluation factor includes: S11. Feature embedding representation: According to the course interaction length sequence and the course interaction time sequence of learner u, construct a course embedding matrix to represent all courses in the interaction sequence, construct a relative course time interval embedding matrix to represent the time interval information between courses, and construct an embedding matrix of course positions to represent the sequential information in the course sequence; where represents the nth element in the course interaction length sequence, Denote the nth element in the course interaction time series; wherein, by converting the learner's interaction sequence into a fixed-length sequence , similarly convert the timestamp sequence into a fixed-length time series .

[0024] S12, construct the time interval perception module: Based on the self-attention mechanism, construct the time interval perception module according to the course embedding matrix and the relative course time interval embedding matrix to obtain a representation with relative course time interval features ; wherein, is the hidden dimension, is the attention function set to prevent and from having too high an inner product weight score; through three trainable weight matrices , , convert the course embedding into query, key, and value matrices: , , , , is the trainable weight parameter matrix; S13, construct the sequential position perception module: Based on the self-attention mechanism, construct the sequential position perception module according to the course position embedding matrix to obtain a representation with sequential position features ; where d is the hidden dimension, h is the attention head set to prevent and from having too high an inner product weight score, , is the trainable weight parameter matrix; in this embodiment, the and the are obtained by matrix transformation to get the query and key matrices.

[0025] S14, feature fusion representation: Linearly weight and fuse the course representation with relative time interval features and the representation with sequential position features to obtain the fused course features , and the fused course features are still processed by linear combination. To enhance the linear expression ability of the model, a non-linear transformation needs to be performed and then the final course feature representation is obtained; wherein, ; in other embodiments, the , or the said .

[0026] S2. Construct a curriculum diversification recommendation model based on capsule network technology. Design a multi-interest extraction layer based on the dynamic routing mechanism, adaptively extract multiple interest vectors from the learner's historical interaction sequence, and accurately depict the diversified interests of learners in different learning scenarios by dynamically clustering the features of different interaction behaviors. The curriculum diversification recommendation model based on capsule network technology includes: S21. Construct a two-layer capsule network, including low-level capsule i and high-level capsule j: Embedding of the vector representing the curriculum features of The low-level capsule i is used to represent the curriculum sequence features; the high-level capsule j is used to represent different interest curriculum types; where , starting from the low-level capsule i, gradually iteratively calculate the interest representation corresponding to each high-level capsule j. The low-level capsule and the high-level capsule The routing logic between is: where, is a bilinear mapping matrix. The high-level capsule is used to capture the relationship between curriculum embedding and interest; S22. Determine the total input vector candidate vector of the high-level capsule ; S23. Output the high-level capsule. By dynamically adjusting the weights, achieve an effective mapping from low-level features to high-level interests. By introducing a non-linear "squashing" function, which can ensure that short vectors shrink to almost zero length and long vectors shrink to a length slightly below 1, the output of the high-level capsule is realized. The output high-level capsule: for the total input vector candidate vector , through the formula: Generate the high-level capsule output v j ; S24. For the learner , the output of the high-level capsule forms a matrix , which is used for subsequent model training and prediction.

[0027] S3. Train and test the models of step S1 and step S2 to train and adjust the parameters, so as to obtain the finally determined models of step S1 and step S2. Among them, in the model training, the multi-interest representation and interaction sequence features of the learner are utilized, and the binary cross-entropy loss function can be adopted to update the model parameters. The model test evaluates the recommendation effect of the model to ensure its accuracy and generalization ability.

[0028] S4. Use the models of step S1 and step S2 finally determined in step S3 to make diversified course recommendations for users. The actual application scenarios can be selected according to the actual situation, such as online education, enterprise training, intelligent education assistants, personalized learning, education e-commerce, and intelligent learning communities, etc. Specifically, the actual application scenario of this embodiment is the online education scenario.

[0029] Specifically, in this embodiment, in step S11, it further includes: when the length of the course interaction length sequence exceeds the preset length n, only retain the last n courses in the course interaction length sequence; when the length of the course interaction length sequence is less than the preset length n, fill in specific items at the front of the course interaction length sequence until the length of the course interaction length sequence reaches n. Wherein, n is a positive integer, and the specific item is a value with n being 1. In other embodiments, the specific item can be selected according to the actual situation, such as any character of n or a preset value.

[0030] When the length of the course interaction time sequence exceeds the preset length n, only retain the last n courses in the course interaction time sequence; when the length of the course interaction time sequence is less than the preset length n, fill in specific items at the front of the course interaction time sequence until the length of the course interaction time sequence reaches n. Wherein, n is a positive integer, and the specific item is a value with n being 1. In other embodiments, the specific item can be selected according to the actual situation, such as any character of n or a preset value.

[0031] Furthermore, in this embodiment, it further includes the course embedding matrix , where is the dimension of the embedding vector, represents the embedding of the th course in the sequence; The relative course time interval embedding matrix , where , represents the relative time interval between course and course ; The embedding matrix of the course position , where is a learnable position embedding matrix, indicating the embedding of the th course in the sequence.

[0032] Optionally, the non-linear transformation in step S14 includes: S141, inputting the fused course features into a feed-forward neural network combined with a ReLU activation function: where , is the corresponding weight matrix, , is the corresponding bias vector; S142 is further optimized through layer normalization, residual connection, and dropout regularization techniques to address issues such as overfitting, vanishing gradients, and long training times during the non-linear transformation process, obtaining the final feature representation : where is the element-wise product, and are the mean and variance for normalizing the variable, and are the learned scale factor and bias term, is the i-th element of the normalized variable, is a set small constant.

[0033] Specifically, in step S22, the total input vector candidate vector , where is the coupling coefficient determined by the iterative dynamic routing process, representing the association strength between the low-level capsule and the high-level capsule; , where represents the th capsule should be the log prior probability of coupling with the

[0034] Preferably, in the model training of step S3, an interest selection layer based on a hard attention mechanism is introduced, and the argmax operation is used to extract the interest vector that best matches the learner with the target course, and the interest representation most relevant to the target course is selected from the interest vectors generated by the model to optimize the recommendation effect. The model training in step S3 includes: S31, extracting the interest vector that best matches the user with the target course , where is the multi - interest representation matrix of the user, representing the embedding vector of the target course; S32. Set the training samples containing the user interest representation and the target course embedding , and calculate the probability of interaction between the user and the target course by inner product: : S33. Calculate the loss function for model training. During the model training process, we optimize the objective function to maximize the probability corresponding to the learner's actual interaction courses, and use the sample - based softmax method for efficient solution: S34. Based on the loss function, optimize the model parameters in steps S1 and S2.

[0035] Specifically, for the model testing in step S3, during model training, to generate recommendation results for the learner's interests, based on the learner's past learning records, multiple interest vectors are extracted. In model testing, each interest vector can independently aggregate the top courses from the global course pool according to the inner - product similarity of the nearest - neighbor library in the interest extraction module, obtaining candidate items, and then obtain the final recommendation results by maximizing the following function. The model testing in step S3 includes: Measure the quality of the recommendation by calculating the overall interest degree of the user in the recommended courses . The is: where, represents the total interest degree of the user in the recommendation set , is the embedding vector of the candidate course, representing the th interest vector of the learner.

[0036] As a preferred embodiment of the present invention, step S4 includes deploying the models in steps S1 and S2 finally determined by step S3 into the actual application scenario to achieve diversified course recommendations based on the learning course interval time.

[0037] The present invention also provides a diversified course recommendation system based on the interval time of learning courses, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the diversified course recommendation method based on the interval time of learning courses as described in any one of the above are implemented.

[0038] The present invention also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the diversified course recommendation method based on the interval time of learning courses as described in any one of the above are implemented.

[0039] Figure 7 is a schematic diagram of the hardware structure for running the diversified course recommendation method based on the interval time of learning courses provided by an embodiment of the present invention. As Figure 7 shown, this embodiment / computer 6 includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, such as a program for running a computational thinking evaluation method based on the fusion of a BP neural network and multi-source data. When the processor 60 executes the computer program 62, the steps in each of the above embodiments of the computational thinking evaluation method based on the fusion of a BP neural network and multi-source data are implemented. Alternatively, when the processor 60 executes the computer program 62, the functions of each module / unit in each of the above device embodiments are implemented.

[0040] Exemplarily, the computer program 62 can be divided into one or more modules / units, and the one or more modules / units are stored in the memory 61 and executed by the processor 60 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 62 in the computer 6.

[0041] The computer 6 can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer 6 device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art can understand that Figure 7 is only an example of the computer 6, and does not constitute a limitation on the computer 6. It may include more or fewer components than shown in the figure, or combine some components, or different components. For example, the computer 6 may further include input and output devices, network access devices, a bus, etc.

[0042] The so-called processor 60 may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0043] The memory 61 may be an internal storage unit of the computer 6, such as the hard disk or memory of the computer 6. The memory 61 may also be an external storage device of the computer 6, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the terminal device. Further, the memory 61 may also include both the internal storage unit of the computer 6 and the external storage device. The memory 61 is used to store the computer program and other programs and data required by the terminal device. The memory 61 may also be used to temporarily store data that has been output or will be output.

[0044] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0045] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0046] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0047] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal device and method can be implemented in other ways. For example, the device / terminal device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in an electrical, mechanical or other forms.

[0048] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0049] In addition, the functional units in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0050] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0051] The above are only specific application examples of the present invention and do not constitute any limitation to the protection scope of the present invention. In addition to the above embodiments, the present invention can also have other implementation manners. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection required by the present invention.

Claims

1. A diversified course recommendation method based on the interval between learning courses, characterized in that: Includes steps: S1, constructs a course feature model based on the time interval of user interaction courses, including: S11, feature embedding representation: based on the course interaction length sequence of learner u And course interaction time series , construct the course embedding matrix Used to represent all courses in the interaction sequence and construct the relative course time interval embedding matrix Used to represent the time interval information between courses and construct the embedding matrix of course positions Used to represent the order information in the course sequence; represents the nth element in the course interaction length sequence, Represents the nth element in the course interaction time series; S12, constructing a time interval perception module: based on the self-attention mechanism and the course embedding matrix and the relative course time interval embedding matrix Construct a time interval perception module to obtain a representation with relative course time interval characteristics ;in, is a hidden dimension, It is used to prevent and The attention function is set because the weight score of the inner product is too high. , , , , is a trainable weight parameter matrix; S13, constructing a sequential position perception module: based on the self-attention mechanism according to the course position embedding matrix Construct a sequential position-aware module to obtain a representation with sequential position features ; where d is the hidden dimension and h is used to prevent and The attention head is set because the weight score of the inner product is too high. is a trainable weight parameter matrix; S14, feature fusion representation: Representing courses with relative time interval features and representations with sequential positional features , perform linear weighting and fusion to obtain the fused course features , and then perform nonlinear transformation to obtain the final course feature representation ;in, ; S2, build a course diversity recommendation model based on capsule network technology, including: S21, construct a two-layer capsule network, including low-level capsule i and high-level capsule j: course feature representation Embedding of vector The low-level capsule i is used to represent the course sequence characteristics; the high-level capsule j is used to represent different interest course types; , the low-grade capsule And the advanced capsule Routing logic between for: ,in, It is a bilinear mapping matrix high-level capsule used to capture the relationship between course embedding and interest; S22, advanced capsule determination via dynamic routing The total input vector candidate vector ; S23, output advanced capsule: candidate vector for the total input vector , through the formula: , generating high-level capsule output v j ; S24, for learners , the high-level capsule output forms a matrix ; S3, performing model training and model testing on the models of step S1 and step S2 to train and adjust parameters to obtain the finalized models of step S1 and step S2; S4, making diversified course recommendations to the user through the models of step S1 and step S2 finally determined in step S3.

2. A diversified course recommendation method based on the interval between learning courses according to claim 1, characterized in that: The step S11 also includes: when the course interaction length sequence When the length of exceeds the preset length n, only the most recent n courses in the course interaction length sequence are retained; when the course interaction length sequence When the length of the course interaction length sequence is less than the preset length n, Fill the front of the specific item until the course interaction length sequence The length of reaches n; wherein n is a positive integer, and the specific item is a value where n is 1; When the course interacts with the time series When the length of exceeds the preset length n, only the most recent n courses in the course interaction time series are retained; when the course interaction time series When the length of is less than the preset length n, in the course interaction time sequence The front of the specific item is filled until the course interaction time series The length of reaches n; wherein n is a positive integer, and the specific item is the value of n being 1.

3. A diversified course recommendation method based on the interval between learning courses according to claim 2, characterized in that: The course embedding matrix ,in, is the dimension of the embedding vector, Indicates the first Embedding of courses; The relative course time interval embedding matrix ,in, , indicating the course and courses The relative time interval between The embedding matrix of the course position ,in, is the learnable position embedding matrix, Indicates the first Embedding of a course.

4. A diversified course recommendation method based on the interval between learning courses according to claim 3, characterized in that: The nonlinear change in step S14 includes: S141, the integrated course features Input a feedforward neural network with ReLU activation function: in, , is the corresponding weight matrix, , is the corresponding bias vector; S142 is further optimized through layer normalization, residual connection and dropout regularization techniques to obtain the final feature representation : in, is the element-wise product, and are the mean and variance used to standardize the variables, and are the learned scaling factors and bias terms, is the ith element of the standardized variable, is a small constant that is set.

5. The method for recommending diversified courses based on the interval between learning courses according to claim 1, characterized in that: The total input vector candidate vector ,in, is the coupling coefficient determined by the iterative dynamic routing process; ,in, Capsules Capsules Logarithmic prior probability of coupling.

6. A diversified course recommendation method based on the interval between learning courses according to claim 1, characterized in that: The model training in step S3 includes: S31, extract the interest vector that best matches the user and the target course , ,in, is the user's multi-interest representation matrix, Embedding vector representing the target class; S32, setting includes user interest representation and target course embedding The training samples , inner product calculation user Courses with Target Probability of interaction : S33, calculate the loss function of model training: S34, optimizing the model parameters of step S1 and step S2 based on the damage function.

7. A diversified course recommendation method based on the interval between learning courses according to claim 6, characterized in that: The model test in step S3 includes: By calculating the user's overall interest in the recommended courses To measure the quality of the recommendation, for: in, Indicates user Recommended Collection Total interest, is the embedding vector of the candidate course, Representing learners No. interest vector.

8. The method for recommending diversified courses based on the interval between learning courses according to claim 1, characterized in that: The step S4 includes deploying the model of the step S1 and the step S2 finally determined by the step S3 to an actual application scenario, so as to realize diversified course recommendations based on the interval time of learning courses.

9. A diversified course recommendation system based on the interval between learning courses, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for recommending diversified courses based on the interval between learning courses as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for recommending diversified courses based on the interval between learning courses as claimed in any one of claims 1 to 8 are implemented.

Images