Research On Recommendation And Search Filter Of Friends Algorithm In Social Networks

With the rapid development of information technology, people’s daily life isoccupied gradually by social networks, such as electronic contacts, entertainment andonline trading. What merchants and customers concerned is recommendation andsearching. On the one hand, the businessmen try to obtain the user’s preferenceinformation accurately for gaining benefits, and then to recommend them to users andpotential customers; on the other hand, users expand their communities by searchingfriends. However, there are a lot of problems with the current recommendation algorithms.The typical collaborative filtering algorithm cannot solve the cold start or new thingsrecommendation problems. The increasing number of registered users makes the results ofsearching friends miscellaneously, which cannot effectively filter the most related users totarget user. This paper researches into this problem, and improves the algorithm onrecommendation and friends searching.First of all, this paper summarizes the traditional thoughts of current recommendationalgorithm, and its pros and cons. Then trust-distrust-based collaborative filtering algorithmwill be discussed. In the trust model, an initial allocation and propagation algorithm oftrust-distrust value is given by using the users’trust value list, then the final value iscomputed by combining interest similarity of users .Finally, the accuracy and validity ofthe algorithm is measured by the recall and average absolute error.Secondly, in view of the problems of the friends search, this paper proposes analgorithm of calculating the correlation between users by means of relation tree, includingthe three modules of interest, address and common friends. In module, the initial score iscalculated according to the similarity of name at first, and then users’path length is found,according to the relationship trees. Finally, correlation score is computed by attenuationfunction. The module sets different weights between the module, and information can bescreen via integrated score.Finally, this paper designs two simulation experiments. The first experiment analysesthe accuracy and validity between the three algorithm of the traditional collaborativefiltering algorithm, trust-based collaborative filtering algorithm and trust-distrust-based collaborative filtering algorithm under the information environment. These two algorithmsanalyze the degree of accuracy on using relation tree to calculate the similarity of users, bycomparing with renren net and cosine similarity algorithm.