Research And Development Of Personalized Recommendation System Based On Network Structure

As an effective method of solving the information overload problem in the Internet age, personalized recommendation system enables users to access effective information through information filtering. Currently, the effect of personalized recommendation system in practice is unsatisfactory, encountering many problems and challenges. Especially in dealing with the actual user data, the algorithm is always not fast enough to be used in real-time application, and the data sparse also reduces recommended accuracy.This paper improved the collaborative filtering recommendation algorithm by considering the nearest neighbor and directed similarity in Douban network data. Then, the improved algorithms were used to recommend books, movies and music for Douban users. The recommended results are carefully compared and analyzed in terms of three well-known indicators including accuracy, diversity and novelty. It is shown that the nearest neighbor algorithm has much lower computational complexity and the directed similarity algorithm obtains higher accuracy, while all these three algorithms have similar diversity and novelty of the recommended results, by comparing with the traditional collaborative filtering recommendation algorithm.To response the information overload problem, as a good complement to the search engine, personalized recommendation system received widespread attention from the researchers allover the world. In this paper, we improved collaborative filtering algorithm and used Douban network data to test and compare different recommendation algorithms, we found a number of new phenomenon based on social network data content and structure of the network and provide experimental data on the social network for the future research of algorithm. In the future, this study can be expanded by considering the across entry type recommendations, friend factors, and time factors affecting the taste of the user’s interest.