Research On Collaborative Filtering Algorithm In Similarity Calculation Method

With the popularization of Internet and the rapid development ofe-commerce, the information in the network growth trend gradually improve,"information overload" problem immediately. User search in the Internet or electroniccommerce needed information and all kinds of convenient service, also faces lost in alot of information of trouble, such not only won’t find your demand information, willbe a waste of time, energy, etc., with the personalized recommendation systemarises at the historic moment.Personalized recommendation system can be based on user interest preference,purchase information such as browsing history.it can help users to recommend thedemand information, goods and services. Recommendation system is the core ofrecommendation algorithm. Good recommendation algorithm can makerecommendation system work better, to meet the needs of users. Recently, a largenumber of researchers and scholars in the recommendation system made a lot ofresearch for recommendation algorithm technology, to solve the problem ofrecommendation system faces recommend quality problems and recommendationsystem.Collaborative filtering algorithm is one of the most widely used recommendationalgorithm in recommender systems, many researchers in view of the presentstudy the problem of collaborative filtering recommendation algorithm, solve theproblem of data sparseness, cold start data recommended system and so on.Collaborative filtering algorithm in similarity calculation method is based on therelationship between the user and the project as the basis, for example, the user like a project or project is a user likes, the algorithm is mainly the relationship between theuser and the project link mining, but between users and projects and projects alsoexist in other aspects of the link between.Heterogeneous information networks in thereal world is everywhere, as the research development of heterogeneous informationnetwork, based on the heterogeneous information network can solve the problem suchas clustering, classification and similarity research, many effective algorithms basedon heterogeneous information networks also gradually presented by research scholars,the study of heterogeneous information network can be found between the objectswithin the territory of the relevant link relations, the use of its rich semanticinformation can be used in study of mining, based on the recommendation of aheterogeneous information network technology application in the field ofrecommendation system will be a new research direction and a valuable research.The main work of this paper has two aspects, one is the calculation method ofsimilarity in collaborative filtering algorithm to improve the classic, facing is used tosolve the problem of collaborative filtering algorithm classic problem; on the otherhand is to recommend links more mining system, method for measuring similarity ofintroducing heterogeneous information network, as the calculation method asimilarity.This paper introduces the research status of the recommender systems andrecommendation algorithm, analysis of the current recommendation system importantproblem faced by the classic recommendation system, and introduces the classicalgorithm in collaborative filtering recommendation algorithm in the first. Basedon the current research status of the collaborative filtering recommendation algorithm,this pape r proposes to improve the idea of the algorithm, and realized theimproved algorithm, the experiment result has certain effect to improve the quality ofthe recommendation system. The method of heterogeneous information network asthe recommendation technology is introduced into the recommendation system, digdeep link relationship between objects and algorithm to calculate therecommendation system based on heterogeneous information networks in the Top-N neighbor set, recommend project supplement for target users, in order to meet thepersonalized needs of users.In this paper, the main work includes:(1) The first detailed introduces the traditional collaborative filteringalgorithms, and analyzes some of the problems that exist. In view of thetraditional collaborative filtering algorithm of data about data sparseness and userinterest migration issues, this article from the aspects of similarity calculation andscore predicts the corresponding improvement, passing through similarity,joined the time function score predicts weight and interest in the user whether thediscriminant formula of migration. Through the experimental results show that, tosome extent the problems existing in the traditional collaborative filteringalgorithm and improve the recommendation accuracy.(2) For the data in the recommendation system is implicit contact relatedquestions, introduce the algorithm of heterogeneous network information asarecommended technique for recommendation system, the relationship betweeneach object in the recommendation system link mining constitute a heterogeneousinformation network, the information network on the basis of the analysis, theapplication of heterogeneous information network similarity measurementalgorithm, the Top-N neighbors of target user sets, to complete the recommendationsystem is recommended. Through experimental analysis based on therecommendation of a heterogeneous information network technology feasibility, andsummarized based on the recommendation of a heterogeneous information networktechnology and the collaborative filtering recommendation technology, as well as thesimilarities and differences of the results of the analysis are presented.