Course recommendation method and device based on multi-dimensional features, electronic device

Abstract

This application relates to the field of deep learning technology, and discloses a course recommendation method, device, and electronic device based on multi-dimensional features. The method includes: generating a user full feature vector based on collected multi-dimensional user data; generating a course full feature vector based on multi-dimensional course data; and inputting the user full feature vector and the course full feature vector into a fusion model to obtain a course recommendation list output by the fusion model. This method, through the comprehensive fusion of multi-dimensional features, effectively solves the problem that traditional recommendations only reflect users' expressed interests and cannot capture potential learning needs, significantly improving the matching accuracy of course recommendations and reducing the information cocoon phenomenon.

Description

Technical Field

[0001] This application relates to the field of deep learning technology, such as a course recommendation method and apparatus based on multi-dimensional features, and electronic devices. Background Technology

[0002] Currently, the online education sector is experiencing explosive growth in course resources. Users face high decision-making costs when confronted with a massive selection of courses. At the same time, education platforms urgently need to improve user retention, completion rates, and satisfaction through precise course recommendations. There are core needs to accurately capture users' potential learning needs, achieve personalized course matching, and solve the cold start dilemma for new users / new courses.

[0003] To address the aforementioned needs, a course recommendation method based on bidirectional matching of user behavior and course content has been disclosed. This method includes: collecting user behavior data such as historical click records, viewing time, and course completion rate; generating user interest vectors through matrix factorization algorithms; extracting content features such as course tags, keywords, and difficulty levels to generate course feature vectors; calculating the cosine similarity between user interest vectors and course feature vectors; and outputting a recommendation list based on the similarity scores.

[0004] In the process of implementing the embodiments of this disclosure, at least the following problems were found in the related art: While related technologies can achieve preliminary course matching based on surface features, reducing user screening costs to some extent, they rely solely on users' historical behavior and surface-level course content features. They fail to integrate deeper key features such as users' learning goals (e.g., certification, skills improvement) and current knowledge level (e.g., beginner / intermediate / advanced). This results in recommendations that only reflect users' expressed interests, failing to capture potential learning needs, leading to low matching accuracy and a tendency to create "information cocoons." Furthermore, these methods are highly dependent on historical data. For new users without behavioral data or new courses without evaluation data, they cannot generate effective feature vectors, only recommending popular courses or random results. This results in poor cold-start performance, severely impacting user experience and platform conversion efficiency.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0007] This disclosure provides a course recommendation method, apparatus, and electronic device based on multi-dimensional features to improve the accuracy of course recommendations.

[0008] In some embodiments, the course recommendation method based on multi-dimensional features includes: generating a user full feature vector based on collected multi-dimensional user data; generating a course full feature vector based on multi-dimensional course data; and inputting the user full feature vector and the course full feature vector into a fusion model to obtain a course recommendation list output by the fusion model.

[0009] Optionally, a user full feature vector is generated based on the collected multi-dimensional user data, including: collecting multi-dimensional user data; wherein the multi-dimensional user data includes behavioral data, learning target data, and knowledge level data; and extracting behavioral features, learning target features, and knowledge level features from the multi-dimensional user data to generate the user full feature vector.

[0010] Optionally, a full feature vector of the course is generated based on multi-dimensional course data, including: acquiring multi-dimensional course data; wherein the multi-dimensional course data includes content data, difficulty data, and evaluation data; and extracting content features, difficulty features, and evaluation features from the multi-dimensional course data to generate a full feature vector of the course.

[0011] Optionally, the user's full feature vector and the course's full feature vector are input into the fusion model to obtain the course recommendation list output by the fusion model, including: concatenating the user's full feature vector and the course's full feature vector to obtain a concatenated vector; and using the fusion model to perform cross-attention calculation on the concatenated vector to obtain the course recommendation list output by the fusion model.

[0012] Optionally, the concatenated vectors are cross-attentioned using a fusion model to obtain a course recommendation list output by the fusion model. This includes: performing association operations through the multi-head attention layer of the fusion model to capture the local association between the user's full feature vector and the course's full feature vector; outputting the original matching score between the user's full feature vector and the associated course's full feature vector through the feedforward network layer of the fusion model; normalizing the original matching score through the output layer of the fusion model to output a matching score; and sorting the courses according to the matching score to obtain a course recommendation list.

[0013] Optionally, the course recommendation method based on multi-dimensional features further includes: when a new user appears, collecting the learning target features and knowledge level features of the new user; based on the learning target features and knowledge level features of the new user, selecting multiple similar users corresponding to the new user from a preset historical user database; and using the mean of the full feature vectors of the multiple similar users as the full feature vector of the new user.

[0014] Optionally, the course recommendation method based on multi-dimensional features further includes: when a new course appears, obtaining the content features and difficulty features of the new course; selecting multiple similar courses corresponding to the new course from a preset course library based on the content features and difficulty features of the new course; and using the average of the full feature vectors of the multiple similar courses as the full feature vector of the new course.

[0015] Optionally, the course recommendation method based on multi-dimensional features further includes: continuously collecting user behavioral feedback data on the course recommendation list; and updating the parameters of the preset historical user database and the fusion model based on the behavioral feedback data.

[0016] In some embodiments, the course recommendation apparatus based on multi-dimensional features includes: a processor and a memory storing program instructions, the processor being configured to execute the course recommendation method based on multi-dimensional features as described above when the program instructions are executed.

[0017] In some embodiments, the electronic device includes: an electronic device body; and a course recommendation device based on multi-dimensional features, as described above, installed on the electronic device body.

[0018] The course recommendation method, apparatus, and electronic device based on multi-dimensional features provided in this disclosure can achieve the following technical effects: In this embodiment, a user full feature vector is first generated based on multi-dimensional user data, breaking through the limitation of traditional recommendations relying solely on single behavioral data and comprehensively covering users' explicit behaviors, potential needs, and ability levels. Then, a course full feature vector is constructed using multi-dimensional course data, fully characterizing the course's content attributes, learning threshold, and user feedback. Finally, the two types of full feature vectors are input into a fusion model, which outputs a course recommendation list through in-depth mining and correlation calculation of the multi-dimensional features.

[0019] By fully integrating multi-dimensional features, this approach effectively addresses the problem that traditional recommendations only reflect users' expressed interests and fail to capture potential learning needs. It significantly improves the matching accuracy of course recommendations, reduces the information cocoon phenomenon, and lays the foundation for cold start optimization and feedback iteration, thereby helping education platforms improve user completion rates and satisfaction.

[0020] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0021] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1This is a schematic diagram of the implementation environment of the course recommendation method based on multi-dimensional features according to an embodiment of this disclosure; Figure 2 This is a schematic diagram of a course recommendation method based on multi-dimensional features provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of another course recommendation method based on multi-dimensional features provided in this embodiment of the disclosure; Figure 4 This is a schematic diagram of another course recommendation method based on multi-dimensional features provided in this embodiment of the disclosure; Figure 5 This is a schematic diagram of another course recommendation method based on multi-dimensional features provided in this embodiment of the disclosure; Figure 6 This is a schematic diagram of a course recommendation device based on multi-dimensional features provided in an embodiment of this disclosure. Detailed Implementation

[0022] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0023] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0024] Unless otherwise stated, the term "multiple" means two or more.

[0025] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0026] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0027] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.

[0028] With the widespread adoption of online education, users face a vast array of course choices, creating a pressing need for efficient recommendation systems to reduce decision-making costs. Simultaneously, the accuracy of course recommendations significantly impacts user retention and satisfaction. Existing recommendation technologies primarily rely on collaborative filtering (such as user-user collaboration and item-item collaboration), content filtering (such as course tag matching), or deep learning models (such as Wide&Deep and DeepFM). However, these technologies generally suffer from limitations such as single feature dimensions, incomplete user profiles, and weak cold-start handling capabilities.

[0029] This application aims to provide a course recommendation method based on multi-dimensional features. By constructing a full-dimensional user profile that includes user behavior, learning objectives, and knowledge level, and combining it with multi-dimensional course features such as course content, difficulty, and evaluation, the method utilizes a fusion model to improve recommendation accuracy. At the same time, it solves the cold start problem through data transfer learning, ultimately achieving on-demand recommendation and precise matching course recommendation services.

[0030] Figure 1 This is a schematic diagram illustrating the implementation environment of the course recommendation method based on multi-dimensional features according to an embodiment of this disclosure. For example... Figure 1 As shown, the implementation environment may include a historical user database 100, a course database 200, and a processor 600.

[0031] The processor 600 inputs the user's full feature vector and the course's full feature vector into the fusion model, calculates the matching score, sorts the results by score to generate a recommendation list, and outputs it to the client.

[0032] When a new user or a new course appears, the processor 600 matches the features of similar users in the historical user database 100 based on the learning objectives and knowledge level filled in by the new user, and matches the features of similar courses in the course database 200 based on the content and difficulty of the new course, generating an initial feature vector to improve the cold start recommendation effect.

[0033] Combination Figure 2 As shown in the embodiments of this disclosure, a course recommendation method based on multi-dimensional features is provided, including: S201 generates a full feature vector of the user based on the collected multi-dimensional user data.

[0034] S202 generates a full feature vector of the course based on multi-dimensional course data.

[0035] S203: Input the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model.

[0036] The course recommendation method based on multi-dimensional features provided in this disclosure first generates a full user feature vector based on multi-dimensional user data, breaking through the limitation of traditional recommendations that rely solely on single behavioral data, and comprehensively covering users' explicit behaviors, potential needs, and ability levels. Then, a full course feature vector is constructed using multi-dimensional course data, fully characterizing the course's content attributes, learning threshold, and user feedback. Finally, the two types of full feature vectors are input into a fusion model, and the model outputs a course recommendation list through in-depth mining and correlation calculation of multi-dimensional features.

[0037] By fully integrating multi-dimensional features, this approach effectively addresses the problem that traditional recommendations only reflect users' expressed interests and fail to capture potential learning needs. It significantly improves the matching accuracy of course recommendations, reduces the information cocoon phenomenon, and lays the foundation for cold start optimization and feedback iteration, thereby helping education platforms improve user completion rates and satisfaction.

[0038] Optionally, a user full feature vector is generated based on the collected multi-dimensional user data, including: collecting multi-dimensional user data; wherein the multi-dimensional user data includes behavioral data, learning target data, and knowledge level data; and extracting behavioral features, learning target features, and knowledge level features from the multi-dimensional user data to generate the user full feature vector.

[0039] In this embodiment, multi-dimensional user data is first systematically collected. Behavioral data includes explicit data reflecting actual learning preferences, such as user clicks, purchases, viewing time, and course completion rates. Learning goal data is obtained through questionnaires and interactive dialogues to gather information on users' potential needs, such as certification, skills enhancement, and interest cultivation. Knowledge level data is determined based on the accuracy and error types of platform test questions to identify the user's current ability level. These three types of data comprehensively depict the user profile from different dimensions. Then, behavioral features, learning goal features, and knowledge level features are extracted from these data and integrated to generate a complete user feature vector.

[0040] For example, when extracting behavioral features, the system collects data on user clicks, course purchases, average viewing time per course, and completion rate over a 7-day period via the client. This data is used to generate a dynamic behavioral vector to represent the user's behavioral characteristics, with a vector dimension of 20. Specifically, the calculation formula is: Behavioral Vector = w1×C + w2×B + w3×T + w4×Cr. Where w1, w2, w3, and w4 are weighting coefficients, which can be optimized using historical data; for example, w1=0.3 and w2=0.4. C represents the number of clicks; B represents the number of courses purchased; T represents the average viewing time per course; and Cr represents the completion rate.

[0041] When extracting learning objective features, users select objectives through questionnaires or extract keywords through chatbots. The target text is then encoded using the BERT model to generate a 50-dimensional target vector. This 50-dimensional target vector is used to represent the user's learning objective features. Among these, the objective might be "professional skills improvement," "academic exam," or "interest cultivation," and the keyword might be "passing the PMP exam within 3 months."

[0042] When extracting knowledge level features, after the user completes the 5 test questions provided by the platform, the system judges the user's answer performance. Based on the accuracy rate and error type, the user is evaluated as beginner, intermediate or advanced, and the level is mapped to a 3-dimensional horizontal vector. For example, beginner is mapped to a vector of [1, 0, 0].

[0043] By comprehensively collecting data and extracting features, user profiles are no longer limited to historical behavior. This ensures the completeness and accuracy of the user's full feature vector, providing high-quality data support for subsequent feature fusion and recommendation calculation. It solves the problem of low matching degree caused by the single user feature in traditional recommendations from the source, making the recommendation results more in line with the user's personalized learning needs.

[0044] Optionally, a full feature vector of the course is generated based on multi-dimensional course data, including: acquiring multi-dimensional course data; wherein the multi-dimensional course data includes content data, difficulty data, and evaluation data; and extracting content features, difficulty features, and evaluation features from the multi-dimensional course data to generate a full feature vector of the course.

[0045] In this embodiment, multi-dimensional course data is first comprehensively acquired. Content data includes core keywords from the course title, introduction, and objectives, directly reflecting the course's core theme and teaching direction. Difficulty data combines instructor ratings (1-5 stars) with historical user completion rates, objectively reflecting the course's learning threshold. Evaluation data gathers numerous genuine user reviews, reflecting the course's teaching quality and user experience feedback. Subsequently, content features, difficulty features, and evaluation features are selectively extracted from this data and integrated to form a comprehensive course feature vector.

[0046] For example, when extracting content features, first extract keywords from the course title, introduction, and objectives. For instance, you can extract "Python data analysis" and "machine learning" as keywords. Then, use the Word2Vec model to generate a 50-dimensional content vector to represent the content features of the course.

[0047] When extracting difficulty features, the course difficulty level is obtained. This level is marked by the instructor when the course is published. Combined with the historical user completion rate, a 5-dimensional difficulty vector is generated to represent the difficulty features of the course. For example, the vector [0, 0, 1, 0, 0] corresponds to the 3-star difficulty of the course.

[0048] When extracting evaluation features, user reviews are crawled, and positive keywords (such as "practical" and "clear") and negative keywords (such as "too basic" and "slow progress") in the reviews are extracted using an LSTM sentiment analysis model. A 30-dimensional evaluation vector is generated from the keywords to represent the evaluation features of the course.

[0049] By collecting and refining multi-dimensional course data, the system avoids the problem of one-sided features caused by traditional recommendations relying solely on surface-level course labels. The multi-dimensional correspondence between the full feature vector of a course and the full feature vector of a user enables the fusion model to accurately capture the deep relationship between user needs and course attributes. For example, it allows users with "intermediate level + data visualization goals" to be accurately matched with courses of corresponding difficulty and topic, significantly improving recommendation matching accuracy. At the same time, it provides data support for generating initial features of new courses in cold start scenarios.

[0050] Optionally, the user's full feature vector and the course's full feature vector are input into the fusion model to obtain the course recommendation list output by the fusion model, including: concatenating the user's full feature vector and the course's full feature vector to obtain a concatenated vector; and using the fusion model to perform cross-attention calculation on the concatenated vector to obtain the course recommendation list output by the fusion model.

[0051] In this embodiment, the user's full feature vector and the course's full feature vector are first concatenated to form a 158-dimensional concatenated vector containing all core dimensions of the user and the course, ensuring complete input of both types of features. Specifically, the user's 20-dimensional behavior vector, 50-dimensional target vector, and 3-dimensional horizontal vector are concatenated to form a 73-dimensional user full feature vector; the course's 50-dimensional content vector, 5-dimensional difficulty vector, and 30-dimensional evaluation vector are concatenated to form an 85-dimensional course full feature vector. Then, the user's full feature vector and the course's full feature vector are concatenated to form a 158-dimensional concatenated vector, which is then input into the fusion model. The fusion model then performs cross-attention calculations on the concatenated vector, focusing on the key correlation points between user and course features through the model's attention mechanism.

[0052] In this embodiment of the disclosure, the dimensions correspond one-to-one with the course categories. Based on the business needs of online education, all courses are divided into 20 core course categories, such as programming, language learning, professional certification, workplace skills, and interest cultivation. The user's behavior vector is set to 20 dimensions, so that the 20 dimensions of the behavior vector correspond to these 20 categories respectively. The value of each dimension represents the user's interest weight in the corresponding course category. For example, the first dimension is the interest weight of programming and the second dimension is the interest weight of language learning.

[0053] By using behavioral data such as the number of clicks, the number of courses purchased, the average viewing time per course, and the completion rate within 7 days, a weighted formula is used to calculate a value in the range of 0 to 1 for each dimension. The higher the value, the stronger the user's behavioral preference for this type of course. The 20-dimensional vector can present the distribution of user interests across all course categories, avoiding the loss of preference expression due to incomplete category coverage.

[0054] In addition, when setting the dimensions of other features, the learning target features are 50-dimensional. Based on the BERT model, 50 dimensions can fully capture the semantic features of the learning target text, covering all common user learning target types.

[0055] The knowledge level features are 3-dimensional, corresponding one-to-one with the three ability levels of primary, intermediate and advanced. They adopt a one-hot encoding method, and the three dimensions can accurately express the user's ability level. The course content vector is 50-dimensional, matching the feature dimensions of the learning objectives. This ensures that the user's learning objectives are aligned with the feature dimensions of the core course topics, making it easier for the model to capture the semantic relationship between the two.

[0056] The difficulty features are 5 dimensions, each corresponding to a difficulty level from 1 to 5 stars.

[0057] The evaluation features are 30-dimensional and based on the LSTM sentiment analysis model. These 30 dimensions can fully cover the positive and negative core keyword categories in user evaluations (such as clear explanation, practical cases, basic content, slow pace, etc.), and can accurately express the user feedback characteristics of the course.

[0058] The concatenation operation achieves a comprehensive integration of multi-dimensional features between users and courses, avoiding matching biases caused by missing features. Meanwhile, cross-attention calculation breaks through the limitation of traditional cosine similarity, which can only calculate single-dimensional associations, and can deeply explore complex relationships between features of different dimensions, such as the high-dimensional fit between "beginner knowledge level + Python introductory goal" and "2-star difficulty + basic Python content". This deep computing approach upgrades recommendation matching from simple vector comparison to precise association of multi-dimensional features, significantly improving the matching accuracy of course recommendations. It also provides a technical path for continuous improvement of recommendation performance through model optimization, effectively solving the problem of insufficient recommendation accuracy caused by the lack of in-depth feature association in traditional recommendations.

[0059] Optionally, the concatenated vectors are cross-attentioned using a fusion model to obtain a course recommendation list output by the fusion model. This includes: performing association operations through the multi-head attention layer of the fusion model to capture the local association between the user's full feature vector and the course's full feature vector; outputting the original matching score between the user's full feature vector and the associated course's full feature vector through the feedforward network layer of the fusion model; normalizing the original matching score through the output layer of the fusion model to output a matching score; and sorting the courses according to the matching score to obtain a course recommendation list.

[0060] In this embodiment, the multi-head attention layer of the fusion model performs association operations on the concatenated vectors simultaneously through eight attention heads, enabling parallel capture of local associations between the user's full feature vector and the course's full feature vector across different dimensions, such as the precise correspondence between user knowledge level and course difficulty, and learning objectives and course content. Subsequently, the feedforward network layer performs deep processing on the associated features captured by the attention layer through two fully connected layers (128 to 64, 64 to 1), outputting a raw matching score that reflects the degree of fit between the two. The output layer normalizes the raw score using the Softmax function, generating a matching score in the 0-1 range to ensure the comparability and reasonableness of the scores. Specifically, the matching score is calculated as follows: y^i = Softmax(W·(User_Emb Course_Emb)+b). Where W is the weight matrix and b is the bias term. This represents the vector dot product. Finally, all courses are sorted according to their matching scores to form a priority-based course recommendation list.

[0061] The multi-head attention mechanism achieves comprehensive capture of multi-dimensional feature associations, avoiding the omission of associations caused by a single attention head, and making the matching calculation more targeted. The deep processing and normalization operations of the feedforward network layer ensure the accuracy and reliability of the matching score. The sorted recommendation list allows users to quickly find courses that best suit their needs. Through this refined calculation process, the matching accuracy of course recommendations is greatly improved, effectively solving the core problem of insufficient matching accuracy in traditional recommendation systems.

[0062] Specifically, the fusion model is a Transformer model. The training process of the Transformer model is as follows: using the user's click or view behavior on the course as a label (click or view is marked as 1, no click or no view is marked as 0), the cross-entropy loss function is used to optimize the model parameters.

[0063] The cross-entropy loss function is: L= (1 / N)∑i=1N [yilog (y^i)+(1 yi) log (1 y^i)] Where L is the cross-entropy loss function, yi is the true label, y^i is the model's predicted matching score, and N is the number of training samples.

[0064] Optionally, the course recommendation method based on multi-dimensional features further includes: when a new user appears, collecting the learning target features and knowledge level features of the new user; based on the learning target features and knowledge level features of the new user, selecting multiple similar users corresponding to the new user from a preset historical user database; and using the mean of the full feature vectors of the multiple similar users as the full feature vector of the new user.

[0065] Combination Figure 3 As shown in the embodiments of this disclosure, another course recommendation method based on multi-dimensional features is provided, including: S301 generates a full feature vector of the user based on the collected multi-dimensional user data.

[0066] S302, when a new user appears, collect the learning objective characteristics and knowledge level characteristics of the new user.

[0067] S303: Based on the new user's learning objectives and knowledge level characteristics, select multiple similar users corresponding to the new user from the preset historical user database.

[0068] S304 uses the mean of the full feature vectors of multiple similar users as the full feature vector of the new user.

[0069] S305 generates a full feature vector of the course based on multi-dimensional course data.

[0070] S306: Input the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model.

[0071] In this embodiment, the learning objective characteristics and knowledge level characteristics of new users are first rapidly collected through questionnaires, interactive dialogues, and other methods. Then, based on these two types of characteristics, multiple similar users with the same learning objectives and knowledge level as the new users are selected from a pre-set historical user database, ensuring that the selected user group has a high degree of homogeneity in needs. Finally, the average of the full feature vectors of these similar users is taken as the initial full feature vector of the new user, providing effective data support for the recommendation calculation of the new user. By using metadata transfer learning, the cold start problem for new users is effectively solved. The generated initial full feature vector accurately reflects the potential needs and ability levels of new users, ensuring that the courses recommended on the first screen of the new user are highly aligned with their own needs, significantly improving the user experience in the cold start scenario.

[0072] Optionally, the course recommendation method based on multi-dimensional features further includes: when a new course appears, obtaining the content features and difficulty features of the new course; selecting multiple similar courses corresponding to the new course from a preset course library based on the content features and difficulty features of the new course; and using the average of the full feature vectors of the multiple similar courses as the full feature vector of the new course.

[0073] Combination Figure 4 As shown in the embodiments of this disclosure, another course recommendation method based on multi-dimensional features is provided, including: S401 generates a full feature vector of the user based on the collected multi-dimensional user data.

[0074] S402 generates a full feature vector of the course based on multi-dimensional course data.

[0075] S403: When a new course is introduced, obtain the content and difficulty characteristics of the new course.

[0076] S404: Based on the content and difficulty characteristics of the new course, select multiple similar courses corresponding to the new course from the preset course library.

[0077] S405, take the average of the full feature vectors of multiple similar courses as the full feature vector of the new course.

[0078] S406: Input the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model.

[0079] In this embodiment, the content and difficulty features of the new course are first extracted. Then, based on these two types of features, multiple similar courses with a content tag overlap of ≥70% and a difficulty difference of ≤1 star are selected from a pre-set course library. This ensures that the selected courses are highly consistent with the new course in core attributes, and that their user rating features are valuable. Finally, the average of the full feature vectors of these similar courses is used as the initial full feature vector of the new course, filling the feature gap caused by the lack of rating data. Through transfer learning of the course's core metadata, the cold start problem of the new course is successfully solved, enabling the new course to accurately reach the target user group, effectively improving the cold start effect of the new course, and helping high-quality new courses quickly gain exposure and recognition.

[0080] Optionally, the course recommendation method based on multi-dimensional features further includes: continuously collecting user behavioral feedback data on the course recommendation list; and updating the parameters of the preset historical user database and the fusion model based on the behavioral feedback data.

[0081] Combination Figure 5 As shown in the embodiments of this disclosure, another course recommendation method based on multi-dimensional features is provided, including: S501 generates a full feature vector of the user based on the collected multi-dimensional user data.

[0082] S502 generates a full feature vector of a course based on multi-dimensional course data.

[0083] S503 inputs the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model.

[0084] S504 continuously collects user feedback data on the course recommendation list.

[0085] S505 updates the parameters of the preset historical user database and fusion model based on behavioral feedback data.

[0086] In this embodiment, user feedback data on recommended courses is continuously collected, including data such as clicks, viewing time, completion rate, favorites, and comments that directly reflect user satisfaction with the recommendation results. Based on this feedback data, the preset historical user database is updated to supplement information such as the latest user behavior characteristics and changes in learning goals, ensuring the dynamic accuracy of user profiles. Furthermore, the parameters of the fusion model are optimized based on the feedback data, adjusting the model's weight allocation for different features, allowing the model to continuously adapt to changes in user needs and updates to course attributes.

[0087] By continuously updating the historical user database, we ensure that the user's full feature vectors can reflect the latest changes in user learning preferences and needs in real time, avoiding recommendation bias caused by fixed user profiles. Through model parameter optimization, the feature association calculation of the fusion model is made more accurate, and the recommendation matching degree is continuously improved. At the same time, this closed-loop mechanism can also provide richer historical data support for cold start optimization, making the selection of similar users and similar courses more accurate, and further improving the cold start effect.

[0088] Combination Figure 6 As shown, this embodiment of the disclosure provides a course recommendation device 60 based on multi-dimensional features, including a processor 600 and a memory 601. Optionally, the device 60 may further include a communication interface 602 and a bus 603. The processor 600, communication interface 602, and memory 601 can communicate with each other via the bus 603. The communication interface 602 can be used for information transmission. The processor 600 can call logical instructions in the memory 601 to execute the course recommendation method based on multi-dimensional features described in the above embodiment.

[0089] Furthermore, the logic instructions in the aforementioned memory 601 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0090] The memory 601, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 600 executes functional applications and data processing by running the program instructions / modules stored in the memory 601, thereby implementing the course recommendation method based on multi-dimensional features in the above embodiments.

[0091] The memory 601 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 601 may include high-speed random access memory and may also include non-volatile memory.

[0092] This disclosure provides an electronic device, including a product body and the aforementioned course recommendation device based on multi-dimensional features. The course recommendation device based on multi-dimensional features is installed on the electronic device body. The installation relationship described herein is not limited to placement within the electronic device body, but also includes installation connections with other components of the electronic device, including but not limited to physical connections, electrical connections, or signal transmission connections. Those skilled in the art will understand that the course recommendation device based on multi-dimensional features can be adapted to feasible electronic device bodies to achieve other feasible embodiments.

[0093] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc., and other media capable of storing program code.

[0094] The foregoing description and accompanying drawings fully illustrate embodiments of this disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in this application means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, the relevant parts can be referred to the description of the method section.

[0095] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0096] The methods and products disclosed in the embodiments herein (including but not limited to devices and equipment) can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may be merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to implement this embodiment according to actual needs. In addition, the functional units in the embodiments of this disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0097] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

Claims

1. A course recommendation method based on multi-dimensional features, characterized in that, include: Generate a full feature vector of the user based on the collected multi-dimensional user data; Generate a full feature vector for the course based on multi-dimensional course data; Input the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model.

2. The method according to claim 1, characterized in that, Based on the collected multi-dimensional user data, a user full feature vector is generated, including: Collect multi-dimensional user data; this multi-dimensional user data includes behavioral data, learning goal data, and knowledge level data. The user's full feature vector is generated by extracting behavioral features, learning goal features, and knowledge level features from multi-dimensional user data.

3. The method according to claim 1, characterized in that, Based on multi-dimensional course data, a full feature vector for the course is generated, including: Obtain multi-dimensional course data; this includes content data, difficulty data, and evaluation data. Content features, difficulty features, and evaluation features are extracted from multi-dimensional course data to generate a full feature vector for the course.

4. The method according to claim 1, characterized in that, Input the user's full feature vector and the course's full feature vector into the fusion model to obtain the course recommendation list output by the fusion model, including: The concatenated vector is obtained by concatenating the user's full feature vector and the course's full feature vector. By using a fusion model to perform cross-attention calculation on the concatenated vectors, a course recommendation list output by the fusion model is obtained.

5. The method according to claim 4, characterized in that, By using a fusion model to perform cross-attention calculation on the concatenated vectors, a course recommendation list output by the fusion model is obtained, including: The association operation is performed through a multi-head attention layer of the fusion model to capture the local association between the user's full feature vector and the course's full feature vector; The feedforward network layer of the fusion model outputs the original matching score of the user's full feature vector and the associated course's full feature vector; The original matching score is normalized by the output layer of the fusion model, and the matching score is output. The courses are sorted according to their matching scores to obtain a recommended course list.

6. The method according to any one of claims 1 to 5, characterized in that, Also includes: When new users are introduced, their learning objectives and knowledge level characteristics are collected. Based on the learning objectives and knowledge level characteristics of new users, multiple similar users corresponding to the new users are selected from the preset historical user database; The mean of the full feature vectors of multiple similar users is used as the full feature vector of the new user.

7. The method according to any one of claims 1 to 5, characterized in that, Also includes: When a new course is introduced, its content and difficulty characteristics should be identified. Based on the content and difficulty characteristics of the new course, select multiple similar courses corresponding to the new course from the pre-set course library; The mean of the full feature vectors of multiple similar courses is used as the full feature vector of the new course.

8. The method according to any one of claims 1 to 5, characterized in that, Also includes: Continuously collect user feedback data on the course recommendation list; The parameters of the preset historical user database and the fusion model are updated based on behavioral feedback data.

9. A course recommendation device based on multi-dimensional features, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute the course recommendation method based on multi-dimensional features as described in any one of claims 1 to 8 when running the program instructions.

10. An electronic device, characterized in that, include: The electronic device itself; The course recommendation device based on multi-dimensional features as described in claim 9 is installed on the electronic device body.

Images