Research On User Profile Model And Personalized Recommendation Of Digital Library

In recent years, along with unceasing renewal of science and technology and quick advancement of human civilization, all aspects of the theory, knowledge and technology have been rapidly developing, and we live in the ocean of information and knowledge. As the concentration and diffusion center of information and knowledge, the status of the library in people’s lives and work is increasingly important. Along with unceasing development of network technology, communication technology, digital storage technology, visualization technology, artificial intelligence technology, etc., all kinds of digital resources are more and more rich, and the digital library based on digital resources has become an important trend in the development of the library in the future. However, despite the development of the digital library like a raging fire, restricted by development level of various software and hardware technologies, as well as human physiological factor, psychological factor, living habits and other factors, the paper books will remain to occupy the dominant position in a fairly long period of time. The transitional form, formed through the compromise and the fusion of the digital library and the traditional library, and at the same time providing digital books and paper books, will be the mainstream form of the digital library in a fairly long period of time.With the sharp increase of the number of library resources, and unceasing speed of people’s rhythm of life and knowledge update, how to enable users to quickly and accurately find interesting content in the vast library resources has already become one of the urgent topics the library science experts and scholars are faced with, and also is an unavoidable problem in existence and development of modern libraries. Data mining technology objectively provides a train of thought on personalized recommendation of the digital library. Personalized recommendation in digital library as the research object, combined with some current emerging technologies such as ontology, concept lattice, association rule, information recommendation, etc., this paper researches and analyzes theory and implementation involved in user modeling, book recommendation and book configuration in the process of personalized recommendation in the digital library. This paper provides a feasible solution to the personalized recommendation of the digital library, and offers a prototype system to verify it. The key issues in this thesis are as follows:(1) User profile model of the digital library Based on Chinese Library Classification (CLC), a book classification ontology is established. As to categories not fully detailed in CLC, new more detailed categories are generated, by concept lattice clustering, for supplement to CLC. On the basis of the book classification ontology, a user profile model for personalized recommendation is established. In view of the insufficiency of CLC ontology in description of interdisciplinary books, a user profile model based on the vector space model is provided. The two user profile models above describe the user’s preference characteristics in common.Some operations and group characteristics of the user profile model are also studied.(2) Book recommendation strategy directly based on the user profile modelThrough the calculation of semantic similarity between the category of the target book in CLC and concepts in CLC ontology, the "nearest neighbors" are obtained. And, according to the scores (user borrowing times) from the "nearest neighbors", possible interest degree of the user to the target book is calculated by weight.(3) Book collaborative recommendation strategy based on the profile model similarityBased on similarity of the user profile model, users closest to the target user are found as "nearest neighbors", who are used to predict the utility of the target book to the target user by collaborative recommendation strategy. Here, the utility of the target book has been calculated.(4) Recommendation strategy for branch librariesAs to unreasonable phenomena on book allocation ratios between branch libraries that generally exist, prediction method on reasonable allocation ratios are studied. For a new book, all users of each branch library are regarded as a "virtual user" as a whole. According to semantic similarity between the CLC category of the target book and concepts in the CLC ontology, the "nearest neighbors" of the CLC category of the target book are found out. Based on borrowing frequency distribution situations of the "nearest neighbors" in each branch library, reasonable allocation ratio of the new book in each branch library is predicted.(5) Book collaborative recommendation strategy based usersFrom the angle of association rules the similarities between users are estimated and the "nearest neighbors" of the target user are found out. And, according to borrowing situations to target books of the "nearest neighbors", interest degree of the target user of the target book is predicted. (6) Applications of concept lattice in the collaborative recommendationThrough analysis of the principles and methods of concept lattice applications in collaborative recommendation, put forward in the existing literatures, combined with service characteristics of digital libraries, the concept lattice is introduced into personalized recommendations in digital libraries, in order to accelerate search speed for "nearest neighbors" during collaborative recommendation and to improve the computational efficiency of collaborative recommendation.(7) Prototype systemA prototype system is presented to verify the theories and algorithms put forward in this paper.