Research On Recommendation Algorithm Based On User's Feedback Information And Heterogeneous Information

With development of internet and information technology,there is a large amount of information posted on the internet at any time,rapidly growing information has made people into the era of "information overload".Whether it is for information consumers or information producers,it is very difficult to find useful information from a large of information or to make your information stand out.Recommendation system is an effective way to solve this problem.The core content of recommendation system is to model user's historical data and predict the possible behavior of users in the future and make relevant recommendations.Collaborative filtering is a most commonly algorithm used for computing similarity in current recommendation system,but there are still some defects,such as data sparsity and scalability issues,which make the final prediction results have limitations.Because the existing algorithm has some defects.This thesis proposes a recommendation algorithm that fuses feedback information and heterogeneous information.The main research works and innovations of this thesis are as follows:1)First of all,this thesis describes the Pearson similarity algorithm in detail.Because Pearson's similarity only considers the scores among users,it ignores that the user's more interested range is usually in an extreme state.With this in mind,this thesis improves Pearson's correlation coefficient and proposes CPSim,which make the user's recommendation more focused on the area of interest to the user.2)Considering these flaws of similarity algorithms based on user ratings,the similarity between users can only be calculated by the common score between users,and the similarity of users without common scores cannot be calculated.Therefore,this thesis combines CPSim with SVD algorithm and uses SVD to process user similarity matrix which obtained by CPSim algorithm,thus solving the problem.3)Solved the problem of algorithm expansion by implementing parallelization calculations of Pathsim,solved the problem of long execution time and memory consumption.Combining feedback information and heterogeneous information,this thesis presents SF-H algorithm.It not only solves the problem that the algorithm based on the heterogeneous information network cannot predict the user's rating and thus cannot be directly applied to the recommendation system.It also solves the problem that collaborative filtering algorithm considers information too single.Experiments with Douban and Yelp datasets have proved that the proposed algorithm has better performance than other well-known recommendation algorithms.In addition,it can effectively solve data sparse and scalability problems.