Dynamic Association Calculated Based Personalized Recommendation System

The rapid development of Internet makes the network data volume increase sharply, it’s an important research topic to help users to obtain needed information quickly and recommend useful information to users. Personalized recommendation service is aimed at the above problems and meets the needs of different users with different backgrounds and predicts the information users interested and recommends to the users based on limited users’behavior.This paper introduces the development of the current main recommendation technology and analyzes the advantages and disadvantages of the content-based and collaborative filtering recommendation algorithm. Then, combining the two recommendation strategy with dynamic correlation calculation to construct the recommendation system model based on dynamic correlation calculation is come up with. This system combines the output of content recommendation based on fruit-plate model and that of collaborative filtering recommendation to produce the final recommendation list and solves the whole system startup problem by a start recommendation list. The recommendation strategy and algorithm description of the whole system and the dynamic correlation calculation parameters settings are given out. In content recommendation based on fruit-plate model, for the recommended term, namely, film that is discussed in this paper, this paper analyzes the characteristics of video data, to construct its content characteristic vector model and the user interest model and proposes to use TF-IDF algorithm to update the user interest model and presents the content recommendation strategy based on fruit-plate model. In the collaborative filtering recommendation, this paper analyzes in details the problems of scarcity and startup and gives out how to solve those problems and makes clear the collaborative filtering recommendation strategy. In the experiment, the Mean Absolute Error, namely MAE, is selected as index to measure recommendation quality. In the same user model and the different system states or the same system state and different user models, the recommendation quality of the system based on dynamic correlation calculation proposed in this paper with the collaborative filtering recommendation system are prepared. The experiments shows that the recommendation quality of the recommendation system based on dynamic correlation calculation, which can provides a better user experience, is better than that of the recommendation system based on collaborative recommendation.