Collaborative Filtering Algorithms Based On Symbolic Data Analysis

With the rapid development of Internet, the number of people that involved in the internet and e-commerce trading platform increased sharply,which result in many e-commerce sites doing researches on the e-commerce technologies.Among the technologies,the collaborative filtering recommendation system is the most widely used. However,due to the information and the enrichment of data, the efficiency and accuracy of traditional collaborative filtering technology is chanllenged.This paper presents a new idea - the symbol data analysis techniques to solve the above problems, using"data package" feature of the symbolic data analysis. Symbol data analysis is the theory and methods that research on how to explore system knowledge from massive amounts of data.There are four main types of symbolic data - -intervalvariable, distributed variable, multi-valued variables, text variables. There are four types of collaborative filtering recommendation algorithms, Item-basedcollaborative filtering(Item-based CF),user-based collaborative filtering (user-basedCF) , model-based collaborative filtering.The general collaborative filtering recommendation algorithm was first introduced, in the following, proposed the direction in the improvement of the existing collaborative filtering algorithms. the item table and user list based on symbolicdata was constructed,based on this , then described how distributed variables and multi-valued variables that being applied in the collaborative filtering algorithm.Based on Item-based CF, user-based CF ,Slope-One CF, which is the three typical traditional collaborative filtering,proposed Item-based symbolic data analysis collaborative filtering (Item-based CFSDA)symbol , user-based symbolic data analysis collaborative filtering(user-based CFSDA),Slope-Onecollaborative Filteringsymbolic data analysis(Slope-One CFSDA) respectively.Finally,the paper selected the data set on the movielens site as empirical example, compared advantages and disadvantages between the proposed method and traditional methods from four parts.The results showed that collaborative filtering algorithm based on the symbolic data analysis was significantly better than traditional recommendation algorithm in both accuracy and recommends efficiency.