Research And Optimization Of User-based Collaborative Filtering Algorithm And System Architecture In Personalized Recommendation System

With the maturity and development of information technology and mobile devices,online services and transactions have become more and more popular,and a lot of information has been gathered to form a burst of data growth.How to solve information overload is an important problem in artificial intelligence and big data age.A great way to solve the problem of information overload is the recommender system.It can help users find new content of interest when the users do not have a specific purpose,and it is an important branch of the field of personalized service research.Collaborative filtering algorithms are recognized as the most well-known and widely used recommendation algorithms,whose core idea is to use the historical behavior data to mine certain similarity to make recommendations.But collaborative filtering algorithms also have some shortages such as cold start,data sparse,lack of accuracy and other key issues.Based on the above background,this thesis has completed the following three aspects of research:1.Optimize the recommender system architecture.This thesis sets up the adaptive feedback mechanism by increasing the feedback link,using the user feedback,context,and group information to adjust the fusion parameters,constituting the system adaptive closed loop.Then the system can be personalized for users to choose the most appropriate recommended algorithm.2.Based on the analysis of a variety of recommendation methods,the collaborative filtering algorithm is studied emphatically and a new collaborative filtering recommendation algorithm based on category and penalty is proposed.,which optimizes the user-based collaborative filtering algorithm.Its main idea is to use a new method based on the combination of column behavior preference and penalty factor to measure user preference similarity.The algorithm effectively utilizes the attribute level,the behavior level,and the hierarchical information.Experiments are conducted based on the MovieLens dataset,and simulation results show that the proposed algorithm greatly improves the evaluation index compared with thecollaborative filtering recommendation algorithm.3.Through the study of natural language processing,this thesis considers combining the latent Dirichlet allocation(LDA)model with the recommender system,and a collaborative filtering recommendation algorithm based on LDA feature extraction is proposed.By learning to generate the corresponding topic and the probability distribution on each topic,it can help to facilitate the establishment of the item feature vector.Finally,using the user similarity calculation method proposed by this thesis that is to add the weight distance factor,generating the nearest neighbor set to complete the recommendation.According to the basic data mining process,the experiment is conducted on the Douban book dataset.Experiment simulation results show that the proposed algorithm realizes the aim of using the text description information to recommend and gets a better recommendation effect.Through the above research,this thesis provides a solid foundation for the further application of the recommender system.