Research On Personalized Recommendation Of University Library Based On Multi-Dimensional Small Data

University library is the guarantee center of university resources and literature service,which undertakes the important task of providing accurate knowledge service for teaching and scientific research.In accordance with the "Education informatization 2.0 action Plan" and other requirements to promote education reform and progress with the use of information technology,in the process of building an open,sharing,interactive and collaborative learning platform,how to use effective technology to provide students with more accurate and personalized knowledge services has crucial study sense and utilization price.In the traditional library service mode based on "book" as the center,Most of the research on personalized recommendation in the field of library and information is resource-centered,using similarity calculation to match people to books,the question of thin data and cool boot cannot be solved properly,and the personalized and precision of recommendation resources are insufficient;The research on library small data is mainly on the theoretical level,but rarely on the practical application of technology.According to the above problem,this paper utilizes the data in constructing multi-dimensional user portrait and design a small data fusion method based on the full connection neural network,using the map neural network framework building book recommendation model,the reconstruction and innovation of library knowledge service model,expand explores the wisdom ability accurate knowledge service of the library.The primary work of this paper below:(1)Research on data Fusion method based on Multi-dimensional Small data User portrait.From the perspective of library small data,Based on the dynamic field theory proposed by Kurt Lewin,a multi-dimensional user portrait is constructed.A data fusion method based on fully connected neural network is designed with user portrait as a bridge,fusion of multi-dimensional small data from the feature level,make the initial embedding include different types of input data to better capture the synergy between the user and the project.Finally,the effectiveness of the proposed method is verified by experimental comparison.(2)Research on library recommendation algorithm based on graph neural network.Based on the fusion of small data,a library recommendation algorithm based on graph neural network is proposed.By transforming user project interaction information and user social information into graph structure,and using the high order connectivity in user project interaction graph and user social graph,we learn the embedded representation of user and item,so as to forecast user hobbies more exactly.Meanwhile,in terms of issue that different graphs of interaction graph and social graph and nodes in the graph have different effects on users' preferences,a multi-level attention mechanism network is designed to learn the influence weights of graph level and node level on user preference.Finally,the effectiveness of the proposed method is verified by experimental comparison.(3)Design and implementation of personalized recommendation system for university library.As to the over commendation algorithm,the commendation system of university library is designed and implemented,the system includes book recommendation,book label,book evaluation,personal space and other functions.Book recommendation is the core module of the system,through the professional attributes in user information and the preference label selected when logging in the system for the first time to solve the problem of cold start commendation system.