Recommendation Algorithm Research Based On Analysis Of User Behavior Data

With the continuous development of network society, the scale of network data also shows exponential growth, so it is hard for users to filter the useless data for themselves from such vast amounts of data in a short time. As a tool to solve the problem of information overload, personalized recommendation technology has been researched and developed during past few years.The main purpose of the personalized recommendation system is allocating resource reasonably with a high efficiency which helps users gain the useful information and data to themselves in a short time from the vast amounts of information. With the increasing of network users, the analysis of user behavior data becomes more and more important in the research of recommendation algorithms,time effect and trust network which coming from user behavior have been hot research field of the recommendation algorithms. Some research achievements about time effect and trust network have been gained in recent years with few disadvantages, yet most research isolated time effect and trust network.So in view of the shortages, we finished the flowing research works from the two aspects of time effect and trust network in this article.(1)We described the basic framework of recommendation system and related research status at home and abroad, introduced three kinds of traditional recommendation algorithms including the recommendation algorithm based on association rules, collaborative filtering recommendation algorithm and hybrid recommendation algorithm, and then analyzed these algorithms to discuss the shortages of these algorithms.(2)We carried on the research works from the two aspects of time effect and trust network, introduced two recommendation algorithms including the recommendation algorithm combining with weighted linear time and the recommendation algorithm based on Beth trust model to summarize these shortages of two algorithms. Then we proposed the recommendation algorithm combining linear weighting method and time window technique and the recommendation algorithm based on the improved Beth trust model. Finally we chose the data set to do the experiment and proved that the improved algorithms have better recommendation results compared with previous algorithms.(3)We proved that trust is also changed over time and designed a new recommendation algorithm combining time effect and improved Beth trust model based on previous improved algorithms, then we chose the data set to do the experiment and proved that the new designed algorithms has better recommendation results compared with previous improved algorithms.