A Book Recommendation System Based On Convolutional Neural Network And Collaborative Filtering Algorithms

With the rapid development of network and big data,people have moved away from the era of information scarcity.The Internet brings convenience to people and provides millions of information,and people have put forward higher requirements for services on the Internet.The explosive growth of information brings two problems.On the one hand,simple search cannot provide personalized services in a targeted manner.It is difficult for users to obtain information they do not know but may be interested in.On the other hand,for a producer of information,how to use information most effectively and push it to those most likely to be interested is also a difficult problem.Nowadays,the books information that provided to users on the Internet is increasing exponentially.Users often need to spend a lot of time browsing and searching for books they may be interested in.In order to optimize the user experience,personalized recommendation algorithm is an effective solution.This thesis proposes a better method for the book recommendation system,and expounds the overall framework and algorithm improvement of the system.The main work of this thesis is as follows:1.The Chinese corpus of books is constructed.Skip-Gram in Word2 Vec is used to model natural language and generate word vectors,comply the preprocessing of convolutional neural network input data.By comparing the influence of three existing corpuses and the books corpus of this thesis on the output of convolution neural network,the conclusion is drawn that the preprocessing result of the books corpus is better than the others.Constructing a convolution neural network.This paper constructs a convolution neural network based on Yoon Kim's,and proves that the convolution neural network in this thesis has excellent results for natural language processing.2.The improved convolution neural network is used as content-based recommendation,and combines with collaborative filtering to improve the recommendation results.The experiment shows that the Hybrid recommendation system has a better result than traditional collaborative filtering and improves the cold start problem to some extent.