The Design And Implementation Of Personlized Recommender System Based On Context Aware And Item Characters

With the growth of Internet information,the types and number of books have further increased,and the way users obtain books is more inclined to use the recommendation system than traditional search and search methods.At present,there are some common problems in the book recommendation system.For example,when the user data is small,the recommendation method is difficult to be effective;the recommendation system performs the book recommendation based on static data such as the user's reading record,without considering the user's dynamic preference and the like.Based on the above shortcomings,this paper designs and implements a book recommendation system.The main research contents include the following four points:(1)Aiming at the problem that the existing book recommendation system considers a single factor in the topic modeling and does not dynamically capture the user preferences over time,a topic model recommendation algoritlhm TCA-LDA that combines time and project features is proposed.The algorithm decomposes the user's reading preferences into topic preferences and time preferences,which together determine which book the user will choose at a certain time.The user's theme preferences are derived from a variety of book characteristics information,such as labels,authors,and publishers,rather than simply from tags,which can effectively alleviate label sparsity.Experiments show that the recommended effect is superior to the existing topic modeling method in recall rate and nDCG.(2)For the sparse problem of modeling data caused by fewer new users,the label-based tensor decomposition recommendation algorithm RSVD is proposed.The algorithm uses the third-order tensor composed of user-book-tags,and considers the weight relationship between different tags.The effect of special tags on books is greater than that of general tags.Experiments show that the algorithm is superior to the traditional tensor decomposition method in accuracy.(3)Focusing on the data sparsity problem of context increasing,a deep learning recommendation algorithm UB-DBN based on comment text is proposed,which uses the review text to construct neural networks for users and books respectively,and design at the high level.The two characteristics of shared network fusion are obtained.Finally,the unknown score is obtained by matrix decomposition.Experiments show that the deep belief network algorithm combined with the comment text can extract the useful information contained in the comments,and has a significant improvement in MAE and RMSE.(4)Design and implement a book recommendation system that integrates the above three research points.The system can provide accurate book recommendation for users according to user needs,and can provide users with adaptive recommendation list in different situations according to the transfer of user needs.The system dynamically crawls and builds user profiles and presents recommendations in a user-friendly manner.