Study On Personalized Book Recommendation Algorithm Based On Reader Behavior Analysis

In recent years,the construction of digital libraries in major universities is in full swing,and in promoting the development of book lending wisdom,how to improve personalized recommendation service for readers among the vast amount of book data is one of the important indicators for judging the quality of library services.Therefore,how to model the diversified lending preferences of users for different characteristics of their offline lending behavior information is the first problem to be solved.Traditional book recommendation algorithms capture users’ current interests from their historical behaviors,which is idealistic for long-term behaviors like borrowing.Therefore,this thesis divides the recommendation work into two stages,recall and sorting,and combines users’ online and offline long-and short-term borrowing behaviors for book recommendation.The main work and innovation points of this thesis are as follows:(1)In the book recall phase,a new diversified serialized book recommendation algorithm model CMINLA-BRM is proposed for the problem of diversified user interests.the model is divided into a multi-interest module and an aggregation module.The multiple interest module captures multiple borrowing behavior preferences from the user’s short-term behavior sequences,and can retrieve candidate sets of items in a large-scale item pool.These items are then fed into the aggregation module to calculate user similarity to obtain an overall predictive score,and the aggregation module uses controllable factors to balance the accuracy and diversity of recommendations.The experiments validate the effectiveness of fusing user offline borrowing behavior sequences to improve recommendation accuracy,and the advantage of the algorithm in tapping diverse user interests.(2)In the book sorting stage,a book recommendation algorithm ULSB-BRM that fuses users’ long-and short-term borrowing behaviors is proposed for long-term borrowing behaviors.The algorithm model uses a self-attentive mechanism to model users’ short-term interests,models users’ long-term interest preferences through collaborative metric learning,and finally uses a gating function similar to an LSTM unit to effectively fuse the two.The experiments demonstrate that the use of a self-attentive mechanism to fuse users’ long-and short-term preferences can effectively mitigate the problem of drifting user interests,and the proposed algorithm has higher recall and accuracy than other related recommendation algorithms.