Research On Collaborative Filtering Recommendation Algorithm Based On Community Detection

With the development of the modern Internet,the information on the Internet becomes more and more complex.The development and progress of the Internet is a double-edged sword.It brings convenience to people's daily work,but also brings information overload.As network information becomes more and more redundant,people often need to extract information from it.In order to dig out more information that users are interested in,many recommendation algorithms have been proposed.Recommendation algorithms are widely used to deal with information filtering problems,they build the interest model from users' historical information,and select appropriate items to recommend to users.Due to the increasing complexity of real networks,recommendation algorithms often have some unnecessary calculations,so the research on recommendation technology has practical significance.In order to reduce the computational complexity of the recommendation algorithm,this paper proposes Recommendation Algorithm based on Community Detection(RACD).The specific work is as follows:(1)In order to improve the traditional item-based collaborative filtering algorithm,Recommendation Algorithm based on Non-Overlapping Community Detection(RANOCD)is proposed.After preprocessing the data,the algorithm divides the item relationship network into communities,i.e.the items with strong associations are classified into a community.Then our algorithm finds items with the user's historical rating higher than 3 points,records the communities they belong to and calculates the user's predicted interest in all items within the communities.Here,a variety of community detection algorithms are used and test on the data with different scales.(2)Overlapping nodes in the networks has a great impact on the recommendation algorithm.This paper proposes Recommendation Algorithm based on Overlapping Community Detection(RAOCD).Taking into account the existence of overlapping nodes in the network,the item relationship network is divided into overlapping communities.A node may be divided into multiple communities at the same time.Then we find items with the user rating higher than 3 points,records the communities they belong to and finally calculate the user's predicted interest in all items within the communities.(3)The data set selected in the experiment is the Movie Lens.In order to obtain accurate research results,data sets of different sizes are selected for training and testing.The results show that the recommendation algorithm based on community detection can effectively reduce the computational redundancy.In the experiment,three indicators of precision,recall and F-score are used to evaluate the performance of the algorithm.Among the RANOCD,the recommendation algorithm based on CLA obtains better performance.We also find that as the list length k increases,RAOCD tends to achieve better performance.