A book recommendation method based on cross-domain learning

Abstract

This invention proposes a book recommendation method based on cross-domain learning, comprising the following steps: acquiring user and book information from a source book domain and at least one target book domain, dividing it into intra-domain book-user interaction data and cross-domain book-category data; constructing an inter-domain book-category relationship graph based on the cross-domain book-category data; constructing an intra-domain book-user interaction graph based on the intra-domain book-user interaction data; performing supervised training on the inter-domain book-category relationship graph to obtain high-order representations of inter-domain book-book categories; combining the high-order representations of inter-domain book-book categories, performing supervised training on the intra-domain book-user interaction graph of each book domain to obtain unified high-order representations of users and books across both domains; predicting the ratings between users in the source book domain and uninteracted books in the target book domain, and recommending potentially interesting books to users in the source book domain in a ranked manner. This invention improves the recommendation accuracy in sparse domains and enhances adaptability to new data.

Description

Technical Field

[0001] This invention relates to the technical field of cross-domain recommendation using deep learning, and more particularly to a book recommendation method. Background Technology

[0002] With the popularization of the internet and the development of information technology, people are faced with a massive amount of information and content, leading to information overload. In this situation, traditional advertising or generalized information push cannot satisfy every user's interests. Users need a personalized recommendation service to help them filter content that matches their interests from a vast library of books. A recommendation system is a technology that uses algorithms to recommend books that users may be interested in. It helps users quickly find books that match their personal preferences from a vast library, providing personalized recommendation services. Recommendation systems typically work based on one or a combination of two methods: content-based recommendation and collaborative filtering. Content-based recommendation records user preferences through browsing history in a single data domain, and then filters the data to find books that are closest to or similar to the user's preferences, greatly improving users' ability to find the books they need online. Through recommendation systems, platforms can better understand user needs and provide more accurate services, thereby increasing business value. However, in most real-world applications, few users can provide ratings or reviews for many books (i.e., data sparsity). This reduces the recommendation accuracy of collaborative filtering models. Almost all existing cross-domain (CF)-based recommender systems are affected to some extent by this long-standing data sparsity problem, especially for new books or new users (the cold start problem). This problem can lead to overfitting when training CF-based models, which significantly reduces recommendation accuracy. Therefore, cross-domain recommendation (CDR) has become a new research topic to improve the data sparsity problem and the cold start problem in the data domain. Summary of the Invention

[0003] To address the data sparsity and cold start problems in existing recommender systems, this invention proposes a book recommendation method based on cross-domain learning, which improves the recommendation accuracy in sparse domains and its adaptability to new data.

[0004] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0005] A book recommendation method based on cross-domain learning includes the following steps:

[0006] Step 1: Obtain user and book information from the source book domain and at least one target book domain, and divide the obtained user and book information into intra-domain book-user interaction data and cross-domain book-category data;

[0007] Step 2: Construct an inter-domain book-category relationship graph based on cross-domain book-category data; construct an intra-domain book-user interaction graph based on intra-domain book-user interaction data;

[0008] Step 3: Using cross-domain book-category data as supervised data, perform supervised training on the inter-domain book-category relationship graph to obtain high-order representations of inter-domain book-book categories;

[0009] Step 4: Using intra-domain book-user interaction data as supervised data, combined with inter-domain book-book category high-order representations, supervised training is performed on the intra-domain book-user interaction graph of each book domain to obtain unified user high-order representations and unified book high-order representations across the two domains.

[0010] Step 5: Repeat steps 3 to 4 until convergence or the set number of training rounds is reached to obtain the final unified user high-order representation and unified book high-order representation.

[0011] Step 6: Using the final unified user high-order representation and unified book high-order representation, predict the recommendation rating between users in the source book domain and books in the target book domain that have not been interacted with, and rank them to recommend potentially interesting books in the target book domain to users in the source book domain.

[0012] Furthermore, the aforementioned intra-domain book-user interaction data includes intra-domain book-user interaction data from source book domain A. User interaction data between books within the target book domain B and the target book domain B. in, and m represents the book ID, u represents the user ID, and r mu This represents the rating that user u gave to book m;

[0013] The cross-domain book category data is as follows: in, and These represent the book title, book description, and book category name, respectively, with 'c' representing the book category ID.

[0014] Furthermore, the method for supervised training on the inter-domain book-category relationship graph is as follows: based on cross-domain book-category data, the original feature vectors of books and the original feature vectors of book categories are obtained through a large language model;

[0015] The original feature vectors of books and the original feature vectors of book categories are mapped and concatenated to obtain the zero-order representation matrix of books-book categories.

[0016] Feature aggregation is performed on the inter-domain book-category relationship graph to obtain a high-order representation of inter-domain book-book category;

[0017] The loss function for inter-domain book-category relationships is calculated based on the high-order representation of inter-domain books-book categories, and the network parameters are updated using the back gradient propagation algorithm based on the calculation results of the loss function.

[0018] Furthermore, the method for supervised training on the book-user interaction graph within each book domain is as follows:

[0019] Based on domain-specific book-user interaction data Mapping and embedding user IDs and book IDs are performed separately to obtain the zero-order representation matrix of book IDs-user IDs; feature aggregation is then performed on the in-domain book-user interaction graph of the source book domain A to obtain the high-order representation H of in-domain book IDs-user IDs of the source book domain A. A ;

[0020] High-order representation H of intra-domain book ID-user ID based on source book domain A A The high-order predictive representation of the target book domain user ID obtained through the overlapping user interest mapper of the source book domain;

[0021] High-order representation H of intra-domain book ID-user ID based on source book domain A A Inter-domain Books - Higher-order Representation of Book Categories H C Calculate the in-domain book-user interaction loss function using the higher-order prediction representation of the target book domain user ID. The network parameters are updated using the back gradient propagation algorithm based on the loss function calculation results.

[0022] Based on domain-specific book-user interaction data Calculate the higher-order representation H of the book ID-user ID within the target book domain B. B and

[0023] Calculate the cross-domain loss of user interest representation based on the higher-order prediction representation of user ID in the target book domain. The parameters of the overlapping user interest mapper in the source book domain are updated using back gradient propagation.

[0024] Furthermore, the method for calculating the original feature vector of the book is as follows:

[0025]

[0026] in, Let be the original feature vector of the book, and Let LLM(·) be the original feature vector matrix of all books, and LLM(·) denote the feature extraction function of the large language model.

[0027] The method for calculating the original feature vector of the book category is as follows:

[0028]

[0029] in, This is the original feature vector of the book's category name. This is the original feature vector matrix for all book categories.

[0030] Furthermore, the formula for calculating the zero-order representation matrix of the book-book category is as follows:

[0031]

[0032] in, All represent the mapping parameters of the original eigenvector matrix. This represents the zeroth-order representation matrix of a book. Book category zero-order representation matrix, This is the zero-order representation matrix of books and book categories. stack(·) represents concatenating the matrix while keeping the columns unchanged.

[0033] The calculation formula for the higher-order representation of inter-domain book-book category is as follows:

[0034]

[0035] Among them, H C This represents the final aggregated inter-domain book-book category high-order representation, where L represents the graph aggregation layer number, and A... C This is the adjacency matrix of the inter-domain book-category relationship graph. For adjacency matrix A C The calculated normalized adjacency matrix of the graph neural network, Δ (l) Let be a noise matrix, and let Δ be the vector in the i-th row of each noise matrix. (l) i For random noise, random noise Δ (l) i Satisfying the L2 norm

[0036] ||Δ (l) i ||2=∈and Let be the parameters that follow a uniform distribution on the interval [0,1], ⊙ represent the cross product, and sign(·) be the sign function. The i-th order representation of inter-domain book-book category.

[0037] Furthermore, the method for calculating the inter-domain book-category relationship loss function based on the higher-order representation of inter-domain book-book category relationships is as follows:

[0038]

[0039]

[0040] in, V represents the total loss function of the inter-domain book-category relationship graph. C This represents the set of nodes in the inter-domain book-category relationship graph. This represents the loss function between book-book category edges. λ represents the graph contrast learning loss function, and λ represents the weight of the graph contrast learning loss function.

[0041] ln(·) is the logarithmic function, σ(·) is the sigmoid activation function, and exp(·) is the exponential function. Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of book categories obtained from the index, where τ is the contrastive learning temperature coefficient;

[0042] (m,c)∈E C This indicates that in the inter-domain book-category relationship graph, there exists a book m belonging to category c, E C The set of edges representing the inter-domain book-category relationship graph; In This indicates that book m does not belong to category c′ in the inter-domain book-category relationship graph.

[0043] Furthermore, the calculation process for obtaining the zero-order representation matrix of book ID-user ID is as follows:

[0044]

[0045] Where Embedding(·) represents the mapping embedding operation. This represents the zeroth-order representation matrix of the book ID. This represents the zeroth-order representation matrix of the user ID. This represents the zero-order representation matrix of book ID - user ID;

[0046] The formula for calculating the higher-order representation of the book ID-user ID within the source book domain A is as follows:

[0047]

[0048] Among them, H A This represents the higher-order representation of the book ID-user ID within the source book domain A obtained by the final aggregation. A This represents the user interaction graph within the source book domain A. For adjacency matrix A AThe calculated normalized adjacency matrix of the graph neural network.

[0049] Furthermore, the calculation method for the higher-order prediction representation of the target book domain user ID is as follows:

[0050]

[0051] in, H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the user ID obtained from the index, projector A→B (·) represents the overlapping user interest mapping function of the source book domain A. This represents the higher-order predictive representation of the user ID in the target book domain B.

[0052] The aforementioned in-domain book-user interaction loss function The method is as follows:

[0053]

[0054] Among them, h u h represents a unified higher-order representation of the user. m V represents the higher-order representation of the unified book. A This represents the set of nodes in the book-user interaction graph within the source book domain A. H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the book ID obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index;

[0055] (u,m)∈E A This indicates the book-user interaction graph within the domain. There exists an interaction pair (r) between user u and book m that shows the user's interest. um ≥1), E A This represents the set of edges in the in-domain book-user interaction graph of the source book domain A. In This represents the user interaction diagram within the domain's books section. There exists an interaction pair (r) where user u is not interested in book m'. um′ <1).

[0056] Furthermore, the cross-domain loss for calculating user interest representations based on the higher-order prediction representation of the target book domain user ID is described. The method is as follows: freeze all parameters except for the overlapping user interest mapper projector of the source book domain A, and calculate the cross-domain loss of user interest representation.

[0057]

[0058] Where Wasserstein(·) denotes the function for calculating the Wasserstein distance. For the user set in the source book domain A, For the set of users in the target book domain B, This represents the higher-order representation H of the book ID-user ID within the target book domain B. B The higher-order representation of the user ID obtained from the index.

[0059] The beneficial effects of this invention are as follows:

[0060] This invention aims to address the data sparsity and cold start problems inherent in traditional recommender systems. Since not all users leave interaction data (ratings, reviews, etc.) after browsing books, the training and prediction data for recommender systems is insufficient, leading to data sparsity. To improve the accuracy of recommendations in the sparse domain, interaction data from the rich domain is learned and mapped to the sparse domain. Furthermore, traditional recommender systems cannot adapt to new data when new users or books are added to the data domain. Therefore, to solve this problem, this invention proposes a learning method based on inter-domain mapping. By analyzing historical interaction data of users and books in the data domain, it achieves rapid matching of new users or new books, and then displays the N books that best match the user's interests as recommended books for the user to choose from. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of the book recommendation method based on cross-domain learning according to the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] A book recommendation method based on cross-domain learning, such as Figure 1 As shown, the steps include:

[0065] Step 1: Obtain user and book information from the source book domain and at least one target book domain, and divide the obtained user and book information into intra-domain book-user interaction data and cross-domain book-category data.

[0066] In this embodiment, user and book information comes from two book domains, namely Book Domain A and Book Domain B. Book Domain A is the source book domain, and Book Domain B is the target book domain. Book Domain A is Douban Reading, and Book Domain B is Qidian Reading. The data in the two book domains may share and overlap to some extent. Users may have different accounts on different platforms, so the results of the other data domain can be recommended based on the data in one data domain.

[0067] Furthermore, the Pandas library was used to process and format user and book information to obtain domain-specific book-user interaction data. Cross-domain books - category data

[0068] Furthermore, domain-specific book-user interaction data Including in-domain book-user interaction data from source book domain A User interaction data between books within the target book domain B and the target book domain B. And there are Where m represents the book ID, u represents the user ID, and r mu This represents the rating that user u gave to book m.

[0069] Furthermore, regarding cross-domain book-category data have and These represent the book title, book description, and book category name, respectively, with 'c' representing the book category ID.

[0070] Examples of datasets are shown in Tables 1 and 2.

[0071] Table 1

[0072]

[0073]

[0074] Table 2

[0075]

[0076] Step 2: Construct an inter-domain book-category relationship graph based on cross-domain book-category data; construct an intra-domain book-user interaction graph based on intra-domain book-user interaction data.

[0077] Furthermore, based on the in-domain book-user interaction data of the source book domain A Retrieve the collection of books in source book domain A User set of source book domain A Using the set of book-user relationships, construct an intra-domain book-user interaction graph for the source book domain A, with books and users as nodes and book-user relationships as edges. Based on the domain book-user interaction diagram of the source book domain A Construct adjacency matrix A A ;in, Denotes the set of nodes, E A ={(m,u)} represents the edge set, i.e., the book-user relationship set, and the element A in the adjacency matrix. A [m][u]=r mu User u's rating for book m is r mu The element A in the adjacency matrix A [m][u] = 0 means that user u did not rate book m.

[0078] Furthermore, based on the intra-domain book-user interaction data of the target book domain B... Retrieve the set of books in the target book domain B. User set of target book domain B Using the set of book-user relationships, construct an intra-domain book-user interaction graph for the target book domain B, with books and users as nodes and book-user relationships as edges. Based on the domain book-user interaction graph of target book domain B Construct adjacency matrix A B ;in, Denotes the set of nodes, E B ={(m,u)} represents the edge set, i.e., the book-user relationship set, and the element A in the adjacency matrix. B [m][u]=r mu The user's rating for book m is r. mu The element A in the adjacency matrix B [m][u] = 0 means the user did not give a rating.

[0079] Furthermore, obtain the book category set based on cross-domain book-category data. Based on the book collection of the source book domain A The set of books in the target book domain B Retrieve book collections from two domains Construct an inter-domain book-category relationship graph using books and book categories from two domains as nodes and book-user relationships as edges. Based on the inter-domain book-category relationship diagram Construct adjacency matrix A C ;in, Denotes the set of nodes, E C ={(m,c)} represents the edge set, A C [m][c] = 1 indicates that book m and book category c are related; otherwise, A C [m][c] = 0. The book m and the book category c are irrelevant.

[0080] Step 3: Using cross-domain book-category data as supervised data, perform supervised training on the inter-domain book-category relationship graph to obtain high-order representations of inter-domain book-book categories.

[0081] The method for supervised training on the inter-domain book-category relationship graph is as follows:

[0082] Furthermore, based on cross-domain book-category data, the original feature vectors of books and book categories are obtained through a large language model.

[0083] The method for calculating the original feature vector of a book is as follows:

[0084]

[0085] in, Let be the original feature vector of the book, and Let LLM(·) be the original feature vector matrix of all books, and let LLM(·) denote the feature extraction function of the large language model. The model used is bge-base-zh-v1.5.

[0086] The method for calculating the original feature vector of book categories is as follows:

[0087]

[0088] in, This is the original feature vector of the book's category name. This is the original feature vector matrix for all book categories.

[0089] Furthermore, the original feature vector matrix of books and the original feature vector matrix of book categories are mapped and concatenated to obtain the zero-order representation matrix of books-book categories;

[0090]

[0091] in, All represent the mapping parameters of the original eigenvector matrix. This represents the zeroth-order representation matrix of a book. Book category zero-order representation matrix, Let be the zero-order representation matrix of books and book categories, d be the dimension of the embedding space, d = 64, |·| represents the number of elements in the set, and stack(·) represents concatenating the matrices without changing the columns. The number of rows in the concatenated matrix is ​​equal to the sum of the number of rows in the original matrix.

[0092] Furthermore, in the inter-domain book-category relationship diagram Feature aggregation is performed to obtain a higher-order representation of inter-domain book-book category;

[0093]

[0094] Among them, H C This represents the final aggregated inter-domain book-book category high-order representation, where L represents the set graph aggregation layer. In the implementation, L=3. For adjacency matrix A C The calculated normalized adjacency matrix of the graph neural network is a set. The elements are noise matrices, and the i-th row vector of each noise matrix is ​​Δ. (l) i For random noise, random noise Δ (l) i Satisfying the L2 norm ||Δ (l) i ||2=∈and The parameters are uniformly distributed over the interval [0,1]. ⊙ represents the vector outer product, sign(·) is the sign function, which takes the value 1 when the input scalar is greater than 0, 0 when it is equal to 0, and -1 when it is less than 0; ∈ represents the upper bound of the noise threshold. The i-th order representation of inter-domain book-book category.

[0095] Furthermore, based on the high-order representation H of inter-domain book-book category... C Calculate the loss function for inter-domain book-book category relationships, and update the network parameters, including the mapping parameters of the original feature vector matrix, using the backpropagation algorithm based on the loss function calculation results. and

[0096] The method for calculating the inter-domain book-category relationship loss function is as follows:

[0097]

[0098] in, The total loss function for the inter-domain book-category relationship graph is represented. This represents the loss function between book-book category edges. λ represents the graph contrast learning loss function, and λ represents the weight of the graph contrast learning loss function. Used to learn the edges that exist on the inter-domain book-category relationship graph. To further distinguish between positive and negative sample pairs of book-book category, the weights λ are summed. Gradient propagation and model parameter updates are performed uniformly.

[0099] ln(·) is the logarithmic function, σ(·) is the sigmoid activation function, and exp(·) is the exponential function. Indicates transpose. Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of book categories obtained from the index, where τ represents the contrastive learning temperature coefficient.

[0100] (m,c)∈E C , representing positive sample pairs, indicates the book-category relationship graph between domains. There exists a book m belonging to category c; In Negative sample pairs represent book-category relationship graphs between domains. Book m in the Chinese dictionary does not belong to category c′. The loss is minimized by maximizing positive sample pairs and minimizing negative sample pairs.

[0101] Step 4: Using intra-domain book-user interaction data as supervised data, combined with inter-domain book-book category high-order representations, supervised training is performed on the intra-domain book-user interaction graph of each book domain to obtain unified user high-order representations and unified book high-order representations across the two domains.

[0102] The method for supervised training on the intra-domain book-user interaction graph of each book domain is as follows:

[0103] In this embodiment, book domain A is used as the source domain and book domain B as the target domain for explanation; alternatively, supervised training can be performed using book domain A as the target domain and book domain B as the source domain, with the same method. Figure 1 As shown.

[0104] Furthermore, based on domain-specific book-user interaction data The user ID and book ID are mapped, embedded, and concatenated respectively to obtain the zero-order representation matrix of book ID-user ID; the calculation process is as follows:

[0105]

[0106] Where Embedding(·) represents the mapping embedding operation, This represents the zeroth-order representation matrix of the book ID. This represents the zeroth-order representation matrix of the user ID. This represents the zero-order representation matrix of book ID - user ID.

[0107] Furthermore, in the domain book-user interaction graph Feature aggregation is performed to obtain the higher-order representation of the book ID-user ID within the source book domain A; the calculation formula is:

[0108]

[0109] Among them, H A This represents the higher-order representation of the book ID-user ID within the source book domain A obtained by the final aggregation, where L represents the number of graph aggregation layers. For adjacency matrix A A The calculated normalized adjacency matrix of the graph neural network.

[0110] Furthermore, based on the higher-order representation H of the intra-domain book ID-user ID of the source book domain A... A The high-order prediction representation of the target book domain user ID obtained through the overlapping user interest mapper of the source book domain:

[0111]

[0112] in, H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the user ID obtained from the index, projector A→B (·) represents the overlapping user interest mapping function of the source book domain A. This represents the higher-order prediction representation of the user ID in the target book domain B.

[0113] Furthermore, based on the higher-order representation H of the intra-domain book ID-user ID of the source book domain A... A Inter-domain Books - Higher-order Representation of Book Categories H C Calculate the in-domain book-user interaction loss function using the higher-order prediction representation of the target book domain user ID. The network parameters, specifically the parameters of the embedding mapping module, are updated using the back gradient propagation algorithm based on the loss function calculation results.

[0114] Intradomain Book-User Interaction Loss Function The method is as follows:

[0115]

[0116] in, Let ln(·) represent the book-user interaction loss, which is used to fit the existing user-book interaction behavior. ln(·) is the logarithmic function, and σ(·) is the sigmoid activation function.

[0117] h u h represents a unified higher-order representation of the user. m This represents a higher-order representation of a unified book. H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the book ID obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index;

[0118] (u,m)∈E A , representing positive sample pairs, indicating the book-user interaction graph within the domain. There exists an interaction pair (r) between user u and book m that shows the user's interest. um ≥1), In Negative sample pairs represent book-user interaction graphs within the domain. There exists an interaction pair (r) where user u is not interested in book m'. um′ <1).

[0119] Furthermore, based on domain-specific book-user interaction data Calculate the higher-order representation H of the book ID-user ID within the target book domain B. B The calculation method is the same as that for the source book domain A, and will not be repeated here.

[0120] Furthermore, all parameters of the overlapping user interest mapper projector, except for the source book domain A, are frozen, and the cross-domain loss of user interest representation is calculated.

[0121]

[0122] in, Represents finding vectors to vector A function of Wasserstein distance, This represents the higher-order representation H of the book ID-user ID within the target book domain B. B The higher-order representation of the user ID obtained from the index.

[0123] Furthermore, the parameters of the overlapping user interest mapper in the source book domain A are updated using back gradient propagation.

[0124] Step 5: Repeat steps 3 to 4 until convergence or the set number of training rounds is reached to obtain the final unified user high-order representation and unified book high-order representation.

[0125] Step 6: Using the final unified user high-order representation and unified book high-order representation, predict the recommendation score between the source book domain user and the target book domain non-interactive books, and rank them to recommend potentially interesting books from the target book domain to the source book domain user.

[0126] Predict the recommendation score between users in the source book domain and non-interactive books in the target book domain. The recommendation score is calculated using the following formula:

[0127]

[0128] Where, p um This indicates the recommended score.

[0129] Furthermore, based on the recommendation scores between the user and the books they have not interacted with, N books that the user in the source book domain might be interested in are recommended to the user in the target book domain, sorted from high to low scores.

[0130] This invention breaks away from the traditional single-objective recommendation system model and utilizes a dual-objective cross-domain recommendation method based on inter-domain mapping. By learning from rich user-book interaction data in the source domain, it maps the user's target book domain, which can solve the problems of data sparsity and cold start in traditional models. This allows it to adapt to the needs of rapid social development and iterative updates, fully leverages the role of deep learning, and quickly recommends books that match users' preferences, thereby improving production efficiency and yielding good economic and social benefits.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A book recommendation method based on cross-domain learning, characterized in that, Including the following steps: Step 1: Obtain user and book information from the source book domain and at least one target book domain, and divide the obtained user and book information into intra-domain book-user interaction data and cross-domain book-category data; Step 2: Construct an inter-domain book-category relationship graph based on cross-domain book-category data; construct an intra-domain book-user interaction graph based on intra-domain book-user interaction data; Step 3: Using cross-domain book-category data as supervised data, perform supervised training on the inter-domain book-category relationship graph to obtain high-order representations of inter-domain book-book categories; Step 4: Using intra-domain book-user interaction data as supervised data, combined with inter-domain book-book category high-order representations, supervised training is performed on the intra-domain book-user interaction graph of each book domain to obtain unified user high-order representations and unified book high-order representations across the two domains. Step 5: Repeat steps 3 to 4 until convergence or the set number of training rounds is reached to obtain the final unified user high-order representation and unified book high-order representation. Step 6: Using the final unified user high-order representation and unified book high-order representation, predict the recommendation rating between users in the source book domain and books in the target book domain that have not been interacted with, and rank them to recommend potentially interesting books in the target book domain to users in the source book domain; The method for supervised training on the book-user interaction graph within each book domain is as follows: Based on domain-specific book-user interaction data The user ID and book ID are mapped, embedded, and concatenated respectively to obtain the zero-order representation matrix of book ID-user ID; feature aggregation is then performed on the in-domain book-user interaction graph of the source book domain A. Obtain the higher-order representation H of the book ID-user ID within the source book domain A. A ; High-order representation H of intra-domain book ID-user ID based on source book domain A A The high-order predictive representation of the target book domain user ID obtained through the overlapping user interest mapper of the source book domain; High-order representation H of intra-domain book ID-user ID based on source book domain A A Inter-domain Books - Higher-order Representation of Book Categories H C Calculate the in-domain book-user interaction loss function using the higher-order prediction representation of the target book domain user ID. The network parameters are updated using the back gradient propagation algorithm based on the loss function calculation results. Based on domain-specific book-user interaction data Calculate the higher-order representation H of the book ID-user ID within the target book domain B. B And calculate the cross-domain loss of user interest representation based on the higher-order prediction representation of the target book domain user ID. The parameters of the overlapping user interest mapper in the source book domain are updated using back gradient propagation.

2. The book recommendation method based on cross-domain learning according to claim 1, characterized in that, The aforementioned intra-domain book-user interaction data includes intra-domain book-user interaction data from source book domain A. User interaction data between books within the target book domain B and the target book domain B. in, and m represents the book ID, u represents the user ID, and r mu This represents the rating that user u gave to book m; The cross-domain book category data is as follows: in, and These represent the book title, book description, and book category name, respectively, with 'c' representing the book category ID.

3. The book recommendation method based on cross-domain learning according to claim 2, characterized in that, The method for supervised training on the inter-domain book-category relationship graph is as follows: based on cross-domain book-category data, the original feature vectors of books and the original feature vectors of book categories are obtained through a large language model; The original feature vectors of books and the original feature vectors of book categories are mapped and concatenated to obtain the zero-order representation matrix of books-book categories. Feature aggregation is performed on the inter-domain book-category relationship graph to obtain a high-order representation of inter-domain book-book category; The loss function for inter-domain book-category relationships is calculated based on the high-order representation of inter-domain books-book categories, and the network parameters are updated using the back gradient propagation algorithm based on the calculation results of the loss function.

4. The book recommendation method based on cross-domain learning according to claim 3, characterized in that, The method for calculating the original feature vector of the book is as follows: in, Let be the original feature vector of the book, and This is the original feature vector matrix of all books. LLM(·) represents the feature extraction function for large language models; The method for calculating the original feature vector of the book category is as follows: in, This is the original feature vector of the book's category name. This is the original feature vector matrix for all book categories.

5. The book recommendation method based on cross-domain learning according to claim 4, characterized in that, The formula for calculating the zero-order representation matrix of the book-book category is as follows: in, All represent the mapping parameters of the original eigenvector matrix. This represents the zeroth-order representation matrix of a book. Book category zero-order representation matrix, This is the zero-order representation matrix of books and book categories. stack(·) represents concatenating the matrix while keeping the columns unchanged. The calculation formula for the higher-order representation of inter-domain book-book category is as follows: Among them, H C This represents the final aggregated inter-domain book-book category high-order representation, where L represents the graph aggregation layer number, and A... C This is the adjacency matrix of the inter-domain book-category relationship graph. For adjacency matrix A C The calculated normalized adjacency matrix of the graph neural network, Δ (l) Let be a noise matrix, and let Δ be the vector in the i-th row of each noise matrix. (l) i For random noise, random noise Δ (l) i Satisfying the L2 norm ||Δ (l) i ||2=∈and Let be the parameters that follow a uniform distribution on the interval [0,1], ⊙ represent the cross product, and sign(·) be the sign function. The i-th order representation of inter-domain book-book category.

6. The book recommendation method based on cross-domain learning according to claim 5, characterized in that, The method for calculating the inter-domain book-category relationship loss function based on the higher-order representation of inter-domain book-book category is as follows: in, V represents the total loss function of the inter-domain book-category relationship graph. C This represents the set of nodes in the inter-domain book-category relationship graph. This represents the loss function between book-book category edges. λ represents the graph contrast learning loss function, and λ represents the weight of the graph contrast learning loss function. ln(·) is the logarithmic function, σ(·) is the sigmoid activation function, and exp(·) is the exponential function. Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of book categories obtained from the index, where τ is the contrastive learning temperature coefficient; (m,c)∈E C This indicates that in the inter-domain book-category relationship graph, there exists a book m belonging to category c, E C The set of edges representing the inter-domain book-category relationship graph; In This indicates that book m does not belong to category c′ in the inter-domain book-category relationship graph.

7. The book recommendation method based on cross-domain learning according to any one of claims 2 to 6, characterized in that, The calculation process for obtaining the zero-order representation matrix of book ID-user ID is as follows: Where Embedding(·) represents the mapping embedding operation. This represents the zeroth-order representation matrix of the book ID. This represents the zeroth-order representation matrix of the user ID. This represents the zero-order representation matrix of book ID - user ID; The formula for calculating the higher-order representation of the book ID-user ID within the source book domain A is as follows: Among them, H A This represents the higher-order representation of the book ID-user ID within the source book domain A obtained by the final aggregation. A This represents the user interaction graph within the source book domain A. For adjacency matrix A A The calculated normalized adjacency matrix of the graph neural network.

8. The book recommendation method based on cross-domain learning according to claim 7, characterized in that, The calculation method for the higher-order prediction representation of the target book domain user ID is as follows: in, H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the user ID obtained from the index, projector A→B (·) represents the overlapping user interest mapping function of the source book domain A. This represents the higher-order predictive representation of the user ID in the target book domain B. The aforementioned in-domain book-user interaction loss function The method is as follows: Among them, h u h represents a unified higher-order representation of the user. m V represents the higher-order representation of the unified book. A This represents the set of nodes in the book-user interaction graph within the source book domain A. H represents the higher-order representation H of the book ID-user ID within the source book domain A. A The higher-order representation of the book ID obtained from the index, Represents the higher-order representation H of inter-domain book-book category. C The higher-order representation of books obtained from the index; (u,m)∈E A This indicates the book-user interaction graph within the domain. There exists an interaction pair (r) between user u and book m that shows the user's interest. um ≥1), E A This represents the set of edges in the in-domain book-user interaction graph of the source book domain A. In This represents the user interaction diagram within the domain's books section. There exists an interaction pair (r) where user u is not interested in book m'. um′ <1).

9. The book recommendation method based on cross-domain learning according to claim 8, characterized in that, The method described above calculates the cross-domain loss of user interest representation based on the higher-order prediction representation of the target book domain user ID. The method is as follows: freeze all parameters except for the overlapping user interest mapper projector of the source book domain A, and calculate the cross-domain loss of user interest representation. Where Wasserstein(·) denotes the function for calculating the Wasserstein distance. For the user set in the source book domain A, For the set of users in the target book domain B, This represents the higher-order representation H of the book ID-user ID within the target book domain B. B The higher-order representation of the user ID obtained from the index.

Images