Research On Collaborative Filtering Recommender Algorithm Based On The Fusion Of Trusted Data And User Similarity

As the main way for people to obtain information,Internet not only brings convenience to people's life,but also makes people have to face the problem of information overload.As recommendation system can effectively alleviate information overload,it is widely used in Internet companies.Among them,collaborative filtering recommendation system,as an important branch of recommendation system,has always been the focus of relevant scholars.Because of the collaborative filtering recommendation algorithm to the recommendation accuracy requirement of the system is higher,the realization of the algorithm is usually very complex,at the same time,inevitably exist in the real network environment inflated ratings mislead the user,the algorithm in dealing with the recommended tasks relevant to the user often problems such as poor performance,in order to solve the above problem,this paper proposes a fusion based on trusted data and improved user similarity of Slope one collaborative filtering recommendation algorithm,in this paper,the main work is as follows:1.In user-based collaborative filtering algorithm,although the traditional user similarity can use cosine similarity and PCC to draw,but might ignore the different users on a single project evaluation of the impact of differences of user similarity computing,namely when calculating the similarity of users,users of the project score vector have significant differences,still can get similar and vector as a result,ultimately affect the system accuracy of recommendation.To solve this problem,an improved user similarity algorithm is proposed.In this algorithm,a balance factor is added to the traditional cosine similarity algorithm to calculate the difference of project score among different users.By testing under the data set of Movie Lens,this paper obtained the most appropriate value of the balance factor,and verified the effectiveness of the optimal threshold value based on the balance factor in the algorithm implementation,which effectively alleviated the influence of users' differences in project scores on the calculation results of similarity and obtained a good recommendation result.2.The further study of the fusion of recommendation algorithm based on trusted data,in order to improve the accuracy of the recommendation system under the sametrusted rate,this paper puts forward a kind of trusted data and improved user similarity fusion slope one algorithm,this algorithm includes three steps,first of all,we choose to have a certain number of trusted data,followed by using the improved similarity algorithm calculate the similarity between users,in the end,we will get weighted similarity to the slope one algorithm,get the final recommendation algorithm.After a series of experiments,the final results show that the improved slope one recommendation algorithm proposed in this paper performs better in MAE and RMSE indexes than the traditional slope one algorithm at the same trust rate,and it can also obtain better recommendation performance when the neighbors number is low,which can alleviate the data sparsity of the scoring matrix and has high practical value.