Research On Algorithms About The Diversity And Novelty In Recommendation System

With the development of information technology and Internet,the data in the network grow explosively.The Internet big data record a large number of users'information and behavior track,it also brings the problem of information overload.Under this background,how to find target items and information quickly becomes a challenge to both ordinary users and Internet information providers.Recommendation system,constructing users' interest model based on the history of users' rating on the items,achieving to recommend items to the users' interest,has become the hottest information filtering and retrieval technology in the era of Information.At present,although increasing studies are focused on the recommendation system' the study about the recommendation system still faces many challenges.For example,how to improve the diversity and novelty without sacrificing accuracy.This study focuses on recommendation diversity and novelty.First,we count the similarity between users based on demographics,then,we cluster users and formalize the popularity of item in every class of users.At last,we propose two demographic attribute-based collaborative filtering algorithms.One is called "accuracy-Oriented collaborative filtering improved algorithm",the other is called“diversity-oriented collaborative filtering improved algorithm".Second,we propose a collaborative filtering improved algorithm based on neighbor diversification in which we improve recommendation diversity and novelty by diversifying the user's neighbors and study the impact of neighbor diversification on the recommendation diversity and novelty.Finally,we propose a recommendation algorithm that fuses the difference between users' distribution of preferences through structuring the personality and breadth of interests of user to quantify users' distribution of preferences and considering the difference between distribution of the items' degree of the users' preferences.Experimental results on the movielens show that the proposed algorithm can improve recommendation diversity and novelty without sacrificing accuracy.Recommendation diversity and novelty are important index in measuring the recommendation quality.But most studies neglect the importance of recommendation diversity and novelty.In this paper,from three different angles--integration of demographics,diverse neighbors and degree distribution of items liked by users,we improve recommendation diversity and novelty without sacrificing accuracy based on the improved collaborative filtering algorithm,which is of great significance in both business intelligence and information service under Internet environment.