CN-121980090-A - Library book personalized recommendation method and system based on condition sharpening collaborative filtering

CN121980090ACN 121980090 ACN121980090 ACN 121980090ACN-121980090-A

Abstract

The invention relates to a library book personalized recommendation method and system based on condition sharpening collaborative filtering, and belongs to the technical field of data processing and information recommendation. The method comprises the steps of constructing a user-book interaction matrix according to an obtained history borrowing record of a library user, constructing a collaborative signal matrix, normalizing, carrying out linear weighted fusion to obtain an initial guide matrix, exploratory remodelling the initial guide matrix to obtain a remodeled guide matrix, carrying out fuzzy processing on the user-book interaction matrix to generate a basic preference matrix, carrying out sharpening processing on the basic preference matrix based on the remodeled guide matrix to obtain a book recommendation score matrix, and carrying out personalized recommendation on the user based on the book recommendation score matrix. The method aims at solving the technical problems that the recommending result is easy to be concentrated in the known interest field of the user, so that the information cocoon house effect is generated, and the novelty and the diversity of the recommending result are limited in the prior art.

Inventors

XU TIANWEI
ZHANG WEISONG
ZHOU JUXIANG

Assignees

云南师范大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (7)

1. A library book personalized recommendation method based on conditional sharpening collaborative filtering, which is characterized by comprising the following steps: Step1, constructing a user-book interaction matrix according to the acquired history borrowing records of library users; step2, constructing a collaborative signal matrix based on the user-book interaction matrix, and carrying out linear weighted fusion on the collaborative signal matrix after normalization to obtain an initial guidance matrix, wherein the collaborative signal matrix comprises a third-order collaborative signal matrix and a fifth-order collaborative signal matrix; step3, exploratory remodeling is carried out on the initial guide matrix to obtain a remolded guide matrix; step4, performing fuzzy processing on the user-book interaction matrix to generate a basic preference matrix; step5, sharpening the basic preference matrix based on the remodeled guide matrix to obtain a book recommendation score matrix so as to conduct personalized recommendation to a user based on the book recommendation score matrix.
2. The personalized library book recommendation method based on the collaborative filtering of conditional sharpening according to claim 1, wherein Step2 is specifically: step2.1 construction of a symmetrically normalized user-item adjacency matrix The expression is: ; Wherein, the For the user-book interaction matrix, And Diagonal matrices of user and item degrees, respectively; Step2.2 for the user-item adjacency matrix Truncated singular value decomposition is carried out to obtain low-rank approximate expression, and the expression is: ; Wherein, the And Is in front of The number of singular vector matrices is chosen, Is in front of A diagonal matrix of individual singular values, Is represented by a transposed matrix; Step2.3, calculating a third-order cooperative signal matrix and a fifth-order cooperative signal matrix in a low-dimensional spectrum space by utilizing a matrix combination law, wherein the expression is as follows: ; Wherein, the Is a three-order cooperative signal matrix, Is a five-order cooperative signal matrix; Step2.4, respectively carrying out normalization processing on the third-order cooperative signal matrix and the fifth-order cooperative signal matrix according to the line L1 to unify the signal scale and convert the signal scale into a probability distribution form, and then introducing weighted superparameters And Linear weighted fusion is carried out on the normalized high-order signals to generate an initial guide matrix The expression is: ; Wherein, the Is a normalization operation.
3. The library book personalized recommendation method based on conditional sharpening collaborative filtering according to claim 2, wherein the exploratory remodeling is a nonlinear transformation of the initial guiding matrix, and the expression is: ; Wherein, the In order to provide a matrix of guidance after remodeling, To explore the adjustment factors, for controlling the degree of inhibition of user-known preferences To indicate a mask matrix for locating user original interaction records Is Hadamard product.
4. The personalized library book recommendation method based on the collaborative filtering of conditional sharpening of claim 3, wherein Step4 is specifically: matrix the user-book interactions The user preference thermodynamic diagram is converted into a smooth and continuous value, and the expression is as follows: ; Wherein, the Representing time-lapse The changed smoothed user preference matrix has dimensions of ; Representing continuous evolution time variable with a value range of ; For a normalized article-to-article similarity matrix, Is that Is a matrix of units of (a); for a preset time interval by a numerical ODE solver Integrating to obtain a basic preference matrix , wherein, Is a super parameter for controlling the degree of blurring process.
5. The personalized library book recommendation method based on the collaborative filtering for conditional sharpening of claim 4, wherein Step5 is specifically: Step5.1: training matrix after the remodeling Interaction matrix with the user-book Superposing to generate a final guide matrix The expression is: ; Wherein, the Is a smoothing coefficient, and is used for ensuring that all books with interaction have basic non-zero guide signals; Step5.2 based on the final guide matrix Constructing a sharpening process expression: ; Wherein, the To be over time An evolving book recommendation score matrix, the initial state of the book recommendation score matrix Based on preference matrix Representing a high pass filtering operation for refining the personalized preferences; the method comprises the steps of guiding intensity parameters, and dynamically adjusting influence weights of external high-order signals on a recommendation result generation track; A linear rectification function with leakage for ensuring that the gradient flow is still preserved when the pilot signal is negative; Step5.3, extracting row vectors of each target user based on a final book recommendation score matrix, filtering books which have been borrowed by the users, sorting the rest books according to score values from high to low, selecting Top-K books with highest ranks, generating a final personalized recommendation list, and realizing personalized recommendation of library books.
6. A library book personalized recommendation system based on condition sharpening collaborative filtering, comprising: The data acquisition and preprocessing module is used for acquiring history borrowing records of library users and constructing a user-book interaction matrix; the guidance matrix construction module is used for constructing a guidance matrix for exploring and guiding according to the user-book interaction matrix; the fuzzy processing module is used for carrying out fuzzy processing on the user-book interaction matrix to generate a smoothed basic preference matrix; the condition sharpening module is used for sharpening the basic preference matrix under the guidance of the guide matrix to generate a final book recommendation score matrix; and the recommendation generation and service module is used for generating a recommendation list according to the final book recommendation score matrix.
7. The system for personalized library book recommendation based on collaborative filtering according to claim 6, wherein the guideline matrix construction module comprises: The high-order signal calculation unit is used for calculating a high-order cooperative signal matrix comprising three-order and five-order connections; the signal fusion unit is used for carrying out normalization and weighted fusion on the high-order cooperative signal matrix; The exploratory remodeling unit is used for reducing the signal intensity of books interacted by a user by applying the inhibition factors; And the guide smoothing unit is used for smoothing the remodeled guide matrix.

Description

Library book personalized recommendation method and system based on condition sharpening collaborative filtering Technical Field The invention relates to a library book personalized recommendation method and system based on condition sharpening collaborative filtering, and belongs to the technical field of data processing and information recommendation. Background In modern library information service, a personalized book recommendation system is a key technology for improving the service quality of readers and improving the utilization rate of collection resources. Collaborative filtering (Collaborative Filtering, CF) is one of the most widely used technical paradigms in the field to predict potential reading interests of individual users by analyzing the borrowing behavior of a population. However, existing collaborative filtering recommendation methods, including some advanced models based on Graph Neural Networks (GNNs), have a common inherent drawback in that they tend to recommend books that are popular or highly similar to user history borrowing records. This mechanism is essentially a "closed loop system" that is good at deep mining around "islands of interest" known to users, but is difficult to guide users to find new books that are far apart in the graph structure but that may be in line with their potential interests. The phenomenon causes gradual convergence of recommendation results, so that a user falls into an information cocoon house, the knowledge field of readers is limited, a large number of valuable books with long tails in a library are difficult to effectively find and utilize, and the circulation of collection resources in the library is unbalanced. Therefore, how to break the closed loop, designing a recommendation mechanism capable of actively guiding users to search interests is a core challenge faced by the current library personalized service. Disclosure of Invention The invention aims to provide a library book personalized recommendation method and system based on condition sharpening collaborative filtering, and aims to solve the technical problems that the recommendation result is easily concentrated in the known interest field of a user, so that the information cocoon house effect is generated, and the novelty and diversity of the recommendation result are limited in the prior art. In order to achieve the above purpose, the technical scheme of the invention is that a library book personalized recommendation method based on condition sharpening collaborative filtering, which converts the sharpening process of recommendation score generation from a non-guided reverse process which is only driven by graph structure in-house into a condition sharpening process which is dynamically guided by an external high-order collaborative signal which is carefully designed to encourage exploration, comprises the following steps: Step1, constructing a user-book interaction matrix according to the acquired history borrowing records of library users; step2, constructing a collaborative signal matrix based on the user-book interaction matrix, and carrying out linear weighted fusion on the collaborative signal matrix after normalization to obtain an initial guidance matrix, wherein the collaborative signal matrix comprises a third-order collaborative signal matrix and a fifth-order collaborative signal matrix; step3, exploratory remodeling is carried out on the initial guide matrix to obtain a remolded guide matrix; step4, performing fuzzy processing on the user-book interaction matrix to generate a basic preference matrix; step5, sharpening the basic preference matrix based on the remodeled guide matrix to obtain a book recommendation score matrix so as to conduct personalized recommendation to a user based on the book recommendation score matrix. Optionally, step2 is specifically: step2.1 construction of a symmetrically normalized user-item adjacency matrix The expression is: Wherein, the For the user-book interaction matrix,AndDiagonal matrices of user and item degrees, respectively; Step2.2 for the user-item adjacency matrix Truncated singular value decomposition is carried out to obtain low-rank approximate expression, and the expression is: Wherein, the AndIs in front ofThe number of singular vector matrices is chosen,Is in front ofA diagonal matrix of individual singular values,Is represented by a transposed matrix; Step2.3, calculating a third-order cooperative signal matrix and a fifth-order cooperative signal matrix in a low-dimensional spectrum space by utilizing a matrix combination law, wherein the expression is as follows: Wherein, the Is a three-order cooperative signal matrix,Is a five-order cooperative signal matrix; Step2.4, respectively carrying out normalization processing on the third-order cooperative signal matrix and the fifth-order cooperative signal matrix according to the line L1 to unify the signal scale and convert the signal scale into a probability distribution