Research On Collaborative Filtering Recommendation Algorithm Based On Dynamic Behavior Of Users

With the rapid development of Internet services, the "Information Explosion" and "Information Overload" promote the development of personalized recommendation system in e-commerce, video entertainment service area and so on. Through the analysis of collected user information to build the user interest model, the recommendation system provides personalized service for users according to the corresponding recommendation algorithm. Therefore designing a more accurate recommendation algorithm based on the history of user-item ratings is more and more important. This is not only the research content of every recommendation system, but also the starting point of this paper. The main research content and innovation of this paper are as follows:1. This paper introduces the recommendation system module designing process, evaluation criterion, common problems of the system and common recommendation algorithms. Through the understanding of recommendation system module designing process and the analysis of common algorithms, we can locate areas needing improvements as well as the merits and demerits of traditional algorithms.2. We put forward the dynamic behavior of user item-based collaborative filtering(DIBCF) considering the time factor. In the neighbor model of collaborative filtering algorithm, the item-item similarity is more stable than user-user similarity. The DIBCF adds the time window to allocate high weight to the item-item similarity included in the time window in order to improve its recommendation performances.3. We put forward the recommendation algorithm based on the dynamic behavior of user SVD++(DSVD++) considering the time factor. In the latent factor model of collaborative filtering algorithm, the current SVD++ is more attractive for its implicit feedback factor and the user-item characteristic dimension, but it lacks the time factor. This paper puts forward the shifting function of user rating bias. It can accurately track user’s interest shift and correct the recommendation rate score, so as to improve the recommendation performance.4. This paper linearly combines DIBCF and DSVD++, proposing the collaborative filtering recommendation algorithm based on dynamic behavior of users(DCF). By integrating the advantages of the stability of item-item similarity in DIBCF, the user-item characteristic dimension and user implicit feedback in DSVD++ and the consideration of the time factor effect, the DCF algorithm can improve the accuracy of recommendation.5. We design simulation experiments based on the Netflix dataset, simulating the algorithm DIBCF, DSVD++ and DCF with the MPI parallel computation mechanism. With the experimental verification, the proposed algorithm DCF has better performance than other traditional algorithms, and it can improve the prediction performance of the recommendation.This paper’s theory analysis and simulation experiments prove that the collaborative filtering recommendation algorithm based on dynamic behavior of users proposed in this paper can improve the recommendation of personalized recommendation system.