Collaborative Filtering Recommendation Algorithm Combined With User Interest

With the rapid development of information age,information overload makes it more difficult to obtain useful information.Personalized recommendation system is produced and helps people to quickly and effectively screen information,it can also actively recommend goods and information to some users.As the core of the recommendation algorithm,it has always been the focus of research.The most widely used algorithm-collaborative filtering algorithm can mine user behavior data and analyze user's preference information to produce personalized recommendation.The main problems involve data sparseness and cold start.The traditional collaborative filtering recommendation algorithm is easy to have sparse score matrix and poor recommendation accuracy.Based on this,this paper proposes a collaborative filtering recommendation algorithm which combines the improved slope one algorithm with the traditional algorithm.Using the project popularity and the LDA topic model,we calculate the user's preference to the project;use the user time data and the user trust model to calculate the user trust degree of the combined time information;add the factors such as the project information and the user relationship to the recommended results in the model,and fill in the evaluation based on the user score data.An improvedrecommendation algorithm,W slope-one collaborative filtering recommendation algorithm,is proposed in the sub matrix.Since the change of user relationship can cause changes in user interest,we add the time based interest weight to the process of similarity calculation and recommendation of the project,in order to get the nearest neighbor set of the project,and then achieve the best recommendation.Experimental verification on the Last.fm data set shows that on the sparse dataset,the recommendation accuracy of the W-slope one collaborative filtering recommendation algorithm combined with the user trust relationship and the time series is 7.45% and 7.68% higher than the traditional model respectively,which indicates that the proposed algorithm can avoid the tradition to a considerable extent.The algorithm is sparse and cold start,and improves the recommendation accuracy and recommendation efficiency.