Teaching resource recommendation method based on user behavior and course duration

Abstract

The embodiment of the invention discloses a teaching resource recommendation method based on user behaviors and course duration. The method comprises the steps of obtaining a to-be-recommended specific user and a plurality of courses; for each course, executing the following operations: generating long-term behavior characteristics of the user by using an SIM model; extracting a first duration with the maximum value from difference values from the occurrence time of the plurality of behaviors input into the SIM model accurate search unit to the current time; according to a relation among a current course duration, the first duration and a second duration covered by the short-term behavior characteristics of the user, performing weighted summation on the long-term behavior characteristics and the short-term behavior characteristics to obtain relative behavior characteristics of the user for the current course; predicting the click rate of the user on the current course by using the relative behavior characteristics and the embedded characteristics of the current course; and recommending a plurality of courses to the user according to the click rate of each course. According to the embodiment, the accuracy of course recommendation is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the fields of artificial intelligence and smart education technology, and in particular to a method for recommending teaching resources based on user behavior and course duration. Background Art

[0002] Teaching resources refer to various materials provided for effective teaching, including textbooks, cases, videos, pictures, courseware, etc. In online learning, how to recommend appropriate courses to users from a large amount of teaching resources is an important aspect of improving learning efficiency and giving full play to the advantages of online learning.

[0003] In the prior art, patent application CN116561410A provides a method for recommending course teaching resources, and patent CN117726082A provides a method for recommending teaching resources, both of which fail to take into account the impact of course duration on user learning behavior, and the accuracy of recommendations is limited. Summary of the invention

[0004] The embodiment of the present invention provides a teaching resource recommendation method based on user behavior and course duration to solve the above technical problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for recommending teaching resources based on user behavior and course duration, comprising:

[0006] S110, obtaining a specific user and multiple courses to be recommended;

[0007] S120. Perform operations from S1-1 to S1-3 for each course:

[0008] S1-1, using the SIM model to generate the long-term behavior characteristics of the user; extracting the first duration with the largest value from the difference between the occurrence time of multiple behaviors input to the SIM model accurate search unit and the current time;

[0009] S1-2, according to the relationship between the current course duration, the first duration, and the second duration covered by the short-term behavior characteristics of the user, weighted summing the long-term behavior characteristics and the short-term behavior characteristics is performed to obtain the relative behavior characteristics of the user for the current course;

[0010] S1-3, predicting the click rate of the user on the current course by using the relative behavior feature and the embedded feature of the current course;

[0011] S130: Recommend multiple courses to the user according to the click rate of each course.

[0012] In a second aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0013] one or more processors;

[0014] a memory for storing one or more programs,

[0015] When the one or more programs are executed by the one or more processors, the one or more processors implement the teaching resource recommendation method based on user behavior and course duration described in any embodiment.

[0016] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the teaching resource recommendation method based on user behavior and course duration described in any embodiment.

[0017] In summary, the embodiment of the present invention provides a teaching resource recommendation method based on user behavior and course duration, combined with the SIM click rate prediction model, to analyze students' past click behaviors on different types of resources, predict their current probability of liking various types of resources, and fit individual learning preferences. At the same time, combined with the laws of video learning behavior and time, different weights are assigned to long-term user behavior and short-term user behavior for different course durations to reflect the complex laws between learning behavior and course duration, further improving the accuracy and personalization of course recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 is a flow chart of a teaching resource recommendation method based on user behavior and course duration provided by an embodiment of the present invention;

[0020] Figure 2 is a structural diagram of a click rate prediction model provided by an embodiment of the present invention;

[0021] Figure 3 is an architecture diagram of a teaching resource recommendation system provided by an embodiment of the present invention;

[0022] Figure 4 It is a structural diagram of a double-tower model provided by an embodiment of the present invention;

[0023] Figure 5 It is a structural diagram of a double-tower model after feedback provided by an embodiment of the present invention;

[0024] Figure 6 It is a structural diagram of another double-tower model after feedback provided by an embodiment of the present invention;

[0025] Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0027] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0028] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0029] Figure 1 1 is a flow chart of a teaching resource recommendation method based on user behavior and course duration provided by an embodiment of the present invention. The method is applicable to online learning and is executed by an electronic device. Figure 1 As shown, the method specifically comprises the following steps:

[0030] S110: Obtain a specific user and multiple courses to be recommended.

[0031] The users and courses here can be users and online courses in the online learning platform. This implementation will take a learning user as an example to illustrate the recommendation process. When recommending for multiple users, the method can be executed for each user separately.

[0032] S120. Use a click rate prediction model including a SIM model to predict the click rate of each course by the specific user.

[0033] The click rate here can be the probability of course learning, which refers to the probability that a specific user will learn each course in the future. Figure 2 The click-through rate prediction model shown in the figure is used for prediction. The model includes a course embedding model, a SIM (Sequential Interaction Model) model, a short-term embedding model, a weighted summation layer, and a prediction layer.

[0034] The course embedding model is used to generate embedding features for courses. Optionally, basic information of course resources (such as course name, lecturer, course duration), attribute information (subject label, difficulty level, course format), and context features (such as learning period, current learning progress environment) can be extracted and input into the course embedding model to obtain an embedding representation of the course.

[0035] The SIM model is used to generate the long-term behavior characteristics of users and recommend courses based on the user's historical behavior. It is particularly suitable for capturing the long-term behavior characteristics of users. The user behavior in this embodiment is the user's learning behavior, including users. The basic principle of SIM is to generate personalized recommendations by modeling the user's behavior sequence, mining the temporal sequence and internal correlation between behaviors, and estimating the click rate based on the user's historical behavior sequence.

[0036] Further, combined with Figure 2 , the input of the SIM model is the user's long-term behavior sequence. Optionally, the courses that the user has learned in a long period of time ending at the current time are extracted, and the embedded features of each course are arranged in chronological order, so that the long-term behavior sequence can be obtained. It should be noted that the time in this embodiment represents a time point.

[0037] The SIM model includes GSU (General Search Unit) and ESU (Exact Search Unit). After the user's long-term behavior sequence is input into GSU, GSU ​​extracts the top K courses with the strongest correlation with candidate course A from the long-term behavior sequence in linear time, and arranges their embedding vectors into a new subsequence, where K is much smaller than the original sequence length M; at the same time, the difference between the time when the user studied the course and the current time (i.e., the time difference, a positive value) is added to the embedding vector of each selected course. The added subsequence is input into ESU, and ESU outputs the user's long-term behavior features through attention operation, which represents the user's long-term interests. Since the courses filtered out in GSU have a weak correlation with course A, the corresponding weight in the attention operation is very small, which hardly affects the final calculation result. Therefore, the SIM model can extract long-term behavior features with very little computational overhead.

[0038] The short-term behavior embedding model is used to generate the short-term behavior features of users. The model can adopt structures such as DIN and DIEN, and the input is the user's short-term behavior sequence. Optionally, extract the courses that the user has studied in a short period of time ending at the current time, and arrange the embedding vectors of each course in chronological order to obtain the short-term behavior sequence; after the sequence is processed by the short-term embedding model, the user's short-term behavior features are output, which represent the user's short-term interests.

[0039] Based on the above three basic models, this embodiment performs the following operations for each course in click rate prediction:

[0040] S1-1. Generate the long-term behavior characteristics of the user using the SIM model, and extract the maximum duration from the difference between the occurrence time of multiple behaviors input to the SIM model ESU and the current time. Figure 2 , for a specific user and the current candidate course A, the user's long-term behavior sequence is input into the SIM model to obtain the user's long-term behavior characteristics; at the same time, the duration from the learning time of each course to the current time is extracted from the top-K subsequences of the input ESU, and a maximum value is extracted as the duration corresponding to the long-term behavior characteristics.

[0041] S1-2. Compare the current course duration, the maximum duration extracted by S1-1, and the duration covered by the user's short-term behavior characteristics, and perform weighted summation of the long-term behavior characteristics and the short-term behavior characteristics according to the relationship between the three durations to obtain the user's relative behavior characteristics for the current course.

[0042] This operation corresponds to Figure 2Specifically, since the learning frequency and interest show a "decay-stabilization" pattern in the study of a series of courses, the click-through rate of the course may show a decaying trend over time. Based on the above rules, this embodiment pays more attention to long-term behavioral characteristics when recommending long courses, and pays more attention to short-term behavioral characteristics when recommending short courses. In order to achieve this effect, this embodiment controls the dimensions of the SIM model and the short-term embedding model to make the two behavioral feature dimensions the same, and performs weighted fusion of the two behavioral features according to the length of the candidate courses, which is used to balance the influence of long-term behavior and short-term behavior when matching with candidate courses later.

[0043] Optionally, the weights of the two behavioral characteristics can be set based on experience. For example, when the course duration is greater than 5 weeks, the weights of the long-term behavioral characteristics and the short-term behavioral characteristics are determined as and ,in ; When the course is less than 5 weeks, the weights of long-term behavior characteristics and short-term behavior characteristics are determined as and ,in, .

[0044] In addition, this embodiment also provides a method to determine different weights for each course through the ratio of the three durations, so as to achieve differentiated weight settings for different courses. For the sake of distinction and description, this embodiment refers to the duration corresponding to the long-term behavior feature extracted by S1-1 as the first duration, and the duration covered by the short-term behavior sequence as the second duration. Usually, the first duration is longer than the second duration, but there are some extreme cases where the second duration is longer. In order to take all situations into account, this embodiment records the longer of the first and second durations as , the shorter one is denoted as , let the current course duration be t, then the weight setting method is:

[0045] If the current course duration is less than the shorter duration stated , then the weight of the behavior feature corresponding to the shorter duration is reset to 1, and the weight of the other behavior feature is reset to 0. That is, the current course duration is very short and is completely driven by short-term interests.

[0046] If the current course duration is longer than the stated longer duration , then the weight of the behavior feature corresponding to the shorter duration is reset to 0, and the weight of the other behavior feature is reset to 1. That is, the current course duration is very long and is completely driven by long-term interests.

[0047] If the current course duration t is between two durations and , the weight of the behavior feature corresponding to the shorter duration is reset to , and reset the weight of the behavior feature corresponding to the longer duration to That is, the current course duration is driven by a combination of short-term interests and long-term interests, and the degree of influence of the two behavioral characteristics is calculated based on the coverage or proportion of the course duration on the two behavioral characteristics.

[0048] Regardless of which method is used, after weighted summation using the above weights, a new behavioral feature can be obtained. This behavioral feature is related to the current course as well as the user's long-term and short-term behaviors, reflecting the complex rules between the user's long-term and short-term behaviors and the course length. This embodiment calls this feature the relative behavioral feature of the current user relative to the current course.

[0049] S1-3, using the relative behavior features and the embedded features of the current course, predicting the click rate of the user on the current course. Figure 2 , the relative feature is weighted fused with the embedding feature of the current course and inputted together Figure 2 The prediction layer in , which reduces the dimension through linear or nonlinear operations, and finally outputs the probability that the specific user will learn the current course in the future.

[0050] S130: Recommend multiple courses to the user according to the click rate of each course.

[0051] After performing the above operations for all courses, the probability of the specific user studying each course in the future can be obtained; the courses with the highest probability are recommended to the specific user.

[0052] After performing the operations of S110 to S130 for each user, the course recommendation for each user can be completed.

[0053] An application scenario of the above method is described below. Figure 3 1 is an architecture diagram of a teaching resource recommendation system provided by an embodiment of the present invention. As shown in FIG. 1 , the system includes:

[0054] Offline layer: Use large models to pre-process massive amounts of teaching resource data, including cleaning, labeling, and classification, to build a teaching resource knowledge base; at the same time, complete model training tasks, train related models such as multi-way recall and click-through rate prediction based on historical data, and generate various data reports for subsequent analysis and optimization. In this stage, the SIM model is incorporated into the training system, and historical teaching resource browsing and usage behavior data is used to train the SIM model to capture users' long-term and short-term learning resource preference patterns, establish a click-through rate prediction model, and provide a basis for subsequent recommendations.

[0055] Near-line layer: focuses on real-time feature processing. On the one hand, it captures the user's learning behavior characteristics in real time, such as the current browsing resource type, learning time, operation records, etc. On the other hand, it monitors the basic attributes of newly launched teaching resources in real time; on this basis, it performs multi-channel recall operations, and based on dual-tower models such as DSSM, it preliminarily screens out thousands of teaching resources that users may be interested in from the content pool and sends them to the coarse ranking layer. At the same time, it also undertakes the coarse ranking task, using the dual-tower model to improve the efficiency of multi-channel recall combinations and reduce the pressure of subsequent fine ranking.

[0056] Online layer: mainly processes real-time user requests, receives hundreds of resources from the near-line layer after rough sorting, uses the highly accurate SIM model of the fine sorting layer to estimate the user's click rate, screens out dozens of resources that best meet the user's current needs, and then uses the MMR algorithm in the re-ranking layer to re-sort them based on accuracy, diversity, and importance, and then delivers them to users to respond to users' resource acquisition needs.

[0057] The teaching resource recommendation method based on user behavior and course duration provided in this embodiment can be applied to the fine ranking layer of the above system. Figure 1 and Figure 2 , the teaching resource recommendation method of the recommendation system performs the following steps in sequence:

[0058] S10, the multi-channel recall layer uses multiple recall branches to provide a recall course set.

[0059] Optional, multi-way recall strategies include:

[0060] Content-based recall strategies: Popular strategy recall, which is to sort the teaching resources according to their overall popularity of browsing, downloading, and use, and give priority to popular courses; attribute strategy recall, which is to accurately locate them according to attributes such as subjects, knowledge points, and resource formats; new strategy recall, which is to push newly uploaded high-quality resources to learners who are interested in the corresponding fields in a timely manner.

[0061] Collaborative filtering recall: A strategy based on similar user behavior patterns, looking for user groups with similar learning behaviors and preferences, and recommending resources of interest to target users; a knowledge-based association strategy, based on the correlation between resources, if a user browses a course, recommends corresponding exercises and analysis videos.

[0062] Embedding vector recall: This branch introduces dual-tower models such as DSSM, constructs independent tower structures for user features and teaching resource features, decouples the calculations of the two, reduces the computational complexity, and requires less computing power for dot product calculations, ensuring real-time performance, generating user vectors in real time, and completing dot product calculations.

[0063] Taking DSSM as an example, combined with Figure 4The dual-tower model consists of a user tower on the left and a course tower on the right. The user tower includes basic user information (such as age, gender, and region), statistical attributes of the learning group to which they belong (major, subject classification), and the user's past learning behavior sequence (browsed, learned, and collected course resources); the course tower covers basic course resource information (course name, lecturer, course duration), attribute information (subject label, difficulty level, course format), and contextual features (such as learning period, current learning progress environment) can be placed in the user tower as appropriate. The matching layer uses inner product or Cosine similarity to measure the matching degree between users and course resources.

[0064] S20. The coarse ranking layer uses another double-tower model to match each recalled course with the user to obtain a plurality of coarse ranking courses and provide them to the fine ranking layer.

[0065] Coarse sorting is the first stage in the recommendation system. Its main purpose is to quickly screen out a group of relatively promising candidates from a large number of candidate sets to reduce the pressure of subsequent calculations. The coarse sorting stage often uses rule-based or simple model-based methods. The coarse sorting layer can also use a double-tower model to construct independent tower structures for user features and teaching resource features, decouple the calculations of the two, reduce the computational complexity, and require less computing power for dot product calculations, ensuring real-time performance, generating user vectors in real time, and completing dot product calculations.

[0066] Finally, the coarse ranking layer adopts a fusion strategy combined with algorithm optimization to select hundreds of resources from the thousands of recalled resources to enter the fine ranking layer, solving the efficiency problem of multi-channel recall combinations and avoiding the performance bottleneck of fine ranking.

[0067] S30, the fine ranking layer uses the multiple roughly ranked courses as the multiple courses to be recommended in S110, executes the above methods of S110-S130, and obtains multiple finely ranked courses.

[0068] Fine sorting refers to a more detailed and accurate sorting of the selected candidates based on rough sorting. This stage often uses complex deep learning models or integrated learning models to pursue higher recommendation accuracy. Figure 3 In this embodiment, based on the click-through rate estimation index, 10 high-quality teaching resources that best match user needs and are most likely to be clicked are strictly selected from the 300 resources sent by the rough ranking and sent to the re-ranking layer. Through the methods of S110-S130, it can be ensured that the screening process fully considers the long-term and short-term dynamic changes of user interest behavior, and avoids the disconnection of recommended resources from the user's current learning process. It should be noted that Figure 3 The number of courses in the table is for reference only and is not an absolute limitation.

[0069] S40, the rearrangement layer executes the MMR algorithm to rearrange the multiple refined courses obtained in S130 to obtain final recommended courses.

[0070] The re-ranking takes into account more business rules, real-time user feedback, and diversity factors, and adjusts and optimizes the results after fine sorting. This embodiment mainly uses the MMR (Maximal Marginal Relevance) algorithm to balance the weights of relevance and diversity, ensuring that the pushed resources are closely related to the user's learning needs, breaking the information cocoon, and improving the long-term use experience. For example, for students who continue to learn mathematics, the core knowledge points are pushed while accompanied by interesting mathematical cases that expand their thinking.

[0071] Furthermore, in order to fully utilize the advantages of the SIM model and relativity as features in capturing the long-term and short-term interests of users, the fine ranking layer of this embodiment can also feed back the behavioral features of each user generated in the SIM model to the dual-tower model of the multi-channel recall layer and / or the coarse ranking layer after executing S130, and fine-tune the dual-tower model after feeding back. The following takes the dual-tower model of the multi-channel recall layer as an example to illustrate the method of feature feeding back and model fine-tuning, which is the same as the dual-tower model of the coarse ranking layer.

[0072] Specifically, regarding the feedback method, according to different feedback levels, this embodiment provides two optional implementation methods:

[0073] The first optional implementation method is to feed back the relative behavioral characteristics of each user for each course to the last layer of the dual tower model. Figure 5 , the feedback process includes the following steps:

[0074] Step 1: The fine ranking layer provides the relative behavioral characteristics of each user for each course to the multi-way recall layer, and the multi-way recall layer adds an MLP (Multilayer Perceptron) between the last layer of the user tower and the matching layer of the double-tower model.

[0075] Step 2: When the dual-tower model is running, the current user's relative behavior characteristics for the current course are combined with the embedding characteristics of the last layer of the user tower (i.e. Figure 5 The concatenated features are input into the multi-layer perceptron for dimension conversion. The output features of the multi-layer perceptron are used to be input into the matching layer together with the last layer of embedding features of the course tower (i.e., the course embedding features) for matching.

[0076] In this implementation, the relative behavior features are fed back to the last layer of the user tower. Before entering the matching layer, the corresponding relative behavior features can be called according to the user and course to be matched to achieve the interaction between the user and the course. The MLP user transforms the concatenated feature dimension to the same dimension as the Item Embedding for Cosine calculation.

[0077] The second optional implementation method is to feed back the relative behavioral characteristics of each user for each course to the first layer of the dual tower model. Figure 6 , the feedback process includes the following steps:

[0078] Step 1: The fine ranking layer uses the relative behavioral characteristics of each user for each course to decouple the independent components of each user and each course in the influence pattern between the user's long-term and short-term behaviors and the course duration, and provides each independent component to the dual-tower model of the multi-channel recall layer.

[0079] Step 2: When the dual-tower model is running, the independent component of each user is fused with its embedded features in the first layer of the user tower, and the independent component of each course is fused with its embedded features in the first layer of the course tower. The two fused features are input into the subsequent processing layers of the user tower and the course tower respectively. The fusion method is not limited, and weighted fusion or dimensionality reduction fusion after splicing can be used.

[0080] The reason for decoupling in this implementation is that the user tower and the course tower in the dual-tower model are independent of each other before the last layer of interaction and do not transmit information to each other. Therefore, the first layer of combination requires decoupling the independent user components and course components from the relative behavioral characteristics.

[0081] In a specific implementation, a user decoupling layer can be pre-constructed. This layer can use a deep learning model to convert the relative behavioral characteristics of each user for multiple courses into the same embedding space. The training of the user decoupling layer can be completed by constraining the conversion results of the same user for multiple courses to be consistent and maximizing the differences in conversion results between different users.

[0082] Furthermore, during the training process, the model can adopt an encoder + decoder architecture, where both the encoder and the decoder use a fully convolutional network or a fully connected layer. After inputting the encoder, the encoded features can be obtained ; Then encode the features Input decoder, you can get the decoding features . The decoding features of the same relative behavior features can be constrained in the loss function Difference from original feature The model parameters are updated by minimizing the encoding features of the relative behavioral features of the same user for multiple courses becoming consistent, and maximizing the differences in the encoding features of the relative behavioral features of different users for each course.

[0083] Exemplarily, the following loss function can be used:

[0084]

[0085] in, represents the loss function value, i, and Represents user index, j, and Respectively represent the course index, , and Represent the corresponding weights respectively.

[0086] After training, the decoder is removed, and the trained encoder is the user decoupling layer. By inputting the relative behavior characteristics of any user for any course into the user decoupling layer, the independent component of the relative behavior characteristics of any user can be obtained, that is, the independent component in the complex law between the user's long-term and short-term behaviors and the course duration.

[0087] Similarly, a course decoupling layer can be pre-constructed. This layer can use a deep learning model to convert the relative behavioral characteristics of multiple users for each course into the same embedding space. The training of the course decoupling layer is completed by constraining the conversion results of multiple users for the same course to be consistent and maximizing the difference in conversion results between different courses.

[0088] Furthermore, during the training process, the model can adopt an encoder + decoder architecture, where both the encoder and the decoder use a fully convolutional network or a fully connected layer. Input the encoder to get the encoding features ; Then encode the features Input decoder, you can get the decoding features . The decoding features of the same relative behavior features can be constrained in the loss function Difference from original feature The model parameters are updated by minimizing the encoding features of the relative behavioral features of multiple users for the same course and maximizing the differences in the encoding features of the relative behavioral features of each user for different courses.

[0089] Exemplarily, the following loss function can be used:

[0090]

[0091] in, represents the loss function value, , and Represent the corresponding weights respectively, and the meanings of other variables are the same .

[0092] After training, the decoder is removed, and the trained encoder is used as the course decoupling layer. By inputting the relative behavior characteristics of any user for any course into the course decoupling layer, the independent component of any course in the relative behavior characteristics can be obtained, that is, the independent component in the complex law of the user's long-term and short-term behavior and the course duration.

[0093] Of the two feedback methods mentioned above, the first optional implementation does not require decoupling, but requires the individual identification of users and courses for each interaction at the matching layer; the second implementation can provide a new user tower that can be reused in various scenarios after one decoupling. In actual applications, it can be flexibly selected according to needs.

[0094] After the feedback is completed, due to the change of the embedding vector, the dual-tower model after the feedback needs to be fine-tuned. This process can be carried out separately outside the user recommendation process, or it can be gradually adjusted in the subsequent course recommendation process. Optionally, this embodiment is based on the principle of reinforcement learning, and the dual-tower model to be fed back is used as the Actor network, the part of the entire recommendation system except the dual-tower model is used as the Critic network, the user is used as the environment, the course actually learned by the user is used as the state variable, the course recommended by the dual-tower model is used as the action variable, and the hit rate of the recommended course actually learned by the user is used as the reward function, and the parameters of the Actor network are fine-tuned using the principle of reinforcement learning. With the accumulation of user data, the Actor network will be gradually optimized, and the network performance can be adjusted in time with the changes in user data.

[0095] It should be noted that the user data involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0096] In summary, the embodiment of the present invention provides a teaching resource recommendation method based on user behavior and course duration, and applies it to a teaching resource recommendation system including a multi-channel recall layer, a coarse ranking layer, a fine ranking layer, and a retake layer, achieving the following beneficial effects:

[0097] 1. The current recommendation system in the education field lacks multiple recall paths and fails to solve the problem of data sparsity. At the same time, the understanding of user characteristics is still at the basic demographic characteristics stage. The user characteristics of the recall strategy are relatively small, which affects the degree of personalization of the recommendation.

[0098] The recommendation method based on multi-channel recall and click-through rate estimation proposed in this embodiment improves the recommendation effect. Through multi-channel recall, resources are recalled from different angles, such as recall based on popular courses, user portraits, knowledge point associations, teacher reputation, learning stages, etc., to ensure the diversity of recommendation results and effectively alleviate the problem of data sparsity. Even if the user interacts less with the system, appropriate recommendation results can be obtained based on multi-dimensional features, ensuring that the resource types are rich and diverse to meet multiple teaching scenarios.

[0099] 2. The current click-through rate prediction model has not fully characterized the learners’ cognitive dynamics and individual learning rules. For example, the existing SIM click-through rate prediction model often ignores the nonlinear rules of learners’ learning behavior and learning time, which makes the prediction of long-term and short-term behavior have systematic deviations, thus affecting the effect of resource recommendation.

[0100] This embodiment combines the SIM click-through rate prediction model to analyze students' past click behaviors on different types of resources, predict their current probability of liking various types of resources, and fit individual learning preferences. More importantly, it combines the rules of video learning behavior and time, and assigns different weights to long-term user behavior and short-term user behavior for different course lengths to reflect the complex rules between learning behavior and course length, further improving the accuracy and personalization of click-through rate prediction.

[0101] 3. The current dual-tower model and SIM model operate independently, and there is a lack of effective information feedback between recall, fine sorting, and coarse sorting, and the information resources between each layer are not fully utilized.

[0102] This embodiment integrates the user behavior characteristics generated based on the SIM model and course duration into the dual-tower model of the recall and coarse sorting strategies, further improving the accuracy of the dual-tower model and realizing the feedback of inter-layer information.

[0103] Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 7 As shown, the device includes a processor 60, a memory 61, an input device 62 and an output device 63; the number of processors 60 in the device can be one or more. Figure 7 A processor 60 is taken as an example; the processor 60, the memory 61, the input device 62 and the output device 63 in the device can be connected by a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0104] The memory 61, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the teaching resource recommendation method based on user behavior and course duration in the embodiment of the present invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 61, that is, realizing the above-mentioned teaching resource recommendation method based on user behavior and course duration.

[0105] The memory 61 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 61 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include a memory remotely arranged relative to the processor 60, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0106] The input device 62 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 63 may include a display device such as a display screen.

[0107] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the teaching resource recommendation method based on user behavior and course duration of any embodiment.

[0108] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0109] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0110] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

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

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A teaching resource recommendation method based on user behavior and course duration, characterized in that: include: S110, obtaining a specific user and multiple courses to be recommended; S120. Perform operations from S1-1 to S1-3 for each course: S1-1, using a SIM model to generate the user's long-term behavior characteristics; Extracting the first duration with the largest value from the differences between the occurrence time of multiple behaviors input to the SIM model accurate search unit and the current time; S1-2, according to the relationship between the current course duration, the first duration, and the second duration covered by the short-term behavior characteristics of the user, weighted summing the long-term behavior characteristics and the short-term behavior characteristics is performed to obtain the relative behavior characteristics of the user for the current course; S1-3, predicting the click rate of the user on the current course by using the relative behavior feature and the embedded feature of the current course; S130: Recommend multiple courses to the user according to the click rate of each course.

2. The method according to claim 1, characterized in that The weighted summing of the long-term behavior feature and the short-term behavior feature according to the relationship between the current course duration, the first duration, and the second duration covered by the short-term behavior feature of the user includes: Determine the longer duration of the first duration and the second duration and shorter duration ; If the current course duration is less than or equal to the shorter duration, the weight of the behavior feature corresponding to the shorter duration is reset to 1, and the weight of the other behavior feature is reset to 0; If the current course duration is greater than or equal to the longer duration, the weight of the behavior feature corresponding to the shorter duration is reset to 0, and the weight of the other behavior feature is reset to 1; If the current course duration t is between the two durations, the weight of the behavior feature corresponding to the shorter duration is reset to , and reset the weight of the behavior feature corresponding to the longer duration to .

3. The method according to claim 1, characterized in that The method is applied to a teaching resource recommendation system including a multi-channel recall layer, a coarse ranking layer and a fine ranking layer, and S110-S130 are performed by the fine ranking layer; Accordingly, before the fine ranking layer executes S110, the method further includes: the multi-path recall layer uses multiple recall branches to provide a recalled course set, wherein the Embedding vector branch uses a double-tower model to provide some recalled courses; the coarse ranking layer uses another double-tower model to match each recalled course with the user, obtains multiple coarse-ranked courses and provides them to the fine ranking layer; Correspondingly, after the fine ranking layer executes S130, the process further includes: the fine ranking layer feeds back the behavioral features of each user to the dual-tower model of the multi-path recall layer and / or the coarse ranking layer, and fine-tunes the dual-tower model after the feedback.

4. The method according to claim 3, characterized in that The fine ranking layer feeds back the behavioral characteristics of each user to the multi-path recall layer and / or the double-tower model of the coarse ranking layer, including: The fine ranking layer provides the relative behavior characteristics of each user for each course to the multi-way recall layer and / or the coarse ranking layer; The multi-way recall layer and / or the coarse sorting layer add a multi-layer perceptron between the last layer of the user tower and the matching layer of the dual-tower model of the current layer; When the dual-tower model is running, the embedding features of the last layer of the user tower are concatenated with the relative behavior features of the current user for the current course and input into the multi-layer perceptron. The output features of the multi-layer perceptron are input into the matching layer to match with the embedding features of the last layer of the course tower.

5. The method according to claim 3, characterized in that: The fine ranking layer feeds back the behavioral characteristics of each user to the multi-path recall layer and / or the double-tower model of the coarse ranking layer, including: The fine ranking layer utilizes the relative behavior characteristics of each user for each course to decouple the independent components of each user and each course in the influence rule between the user's long-term and short-term behaviors and the course duration, and provides each independent component to the dual-tower model of the multi-channel recall layer and / or the coarse ranking layer; When the dual-tower model is running, the independent components of each user are fused with the embedded features of the first layer of the user tower, and the independent components of each course are fused with the embedded features of the first layer of the course tower. The two fused features are input into the subsequent processing layers of the user tower and the course tower respectively.

6. The method according to claim 5, characterized in that The refined ranking layer uses the relative behavioral characteristics of each user for each course to decouple the independent components of each user and each course in the influence rule between the user's long-term and short-term behaviors and the course duration, including: Pre-construct a user decoupling layer, convert the relative behavioral characteristics of each user for multiple courses into the same embedding space, and complete the training of the user decoupling layer by constraining the conversion results of the same user for multiple courses to be consistent and maximizing the difference in conversion results between different users; Pre-constructing a course decoupling layer, converting the relative behavioral characteristics of multiple users for each course into the same embedding space, and completing the training of the course decoupling layer by constraining the conversion results of multiple users for the same course to be consistent and maximizing the difference in conversion results between different courses; The relative behavioral characteristics of any user for any course are input into the trained user decoupling layer and the trained course decoupling layer respectively, and the independent components of any user and any course in the influence rule between the user's long-term and short-term behaviors and the course duration are obtained respectively.

7. The method according to claim 3, characterized in that The fine-tuning of the double-tower model after the feedback includes: The twin-tower model is used as the Actor network, the part of the entire teaching resource recommendation system except the twin-tower model is used as the Critic network, the user is used as the environment, the courses actually studied by the user are used as the state variable, the courses recommended by the twin-tower model are used as the action variable, and the hit rate of the recommended courses actually studied by the user is used as the reward function. The parameters of the Actor network are fine-tuned using the principle of reinforcement learning.

8. The method according to claim 3, characterized in that The teaching resource recommendation system further includes a re-ranking layer; accordingly, after the fine ranking layer executes S130, it further includes: The rearrangement layer executes the maximum margin correlation algorithm to rearrange the multiple courses obtained in S130 to obtain final recommended courses.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the teaching resource recommendation method based on user behavior and course duration as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by a processor, implements the teaching resource recommendation method based on user behavior and course duration as described in any one of claims 1-8.

Images