Research Of Hybrid Recommendation System Based On User Interest And Domain Nearest Neighbor

In the face of the era of big data,how to accurately recommend the information from the vast amount of information in the ocean to the user interested in the information,which will be the main task of the recommendation algorithm research.The most classic two recommendation algorithms are content-based filtering and collaborative filtering recommendation algorithm,but the classic recommendation algorithm also has its own shortcomings.Data sparsity and cold start are the main problems in collaborative filtering recommendation algorithm.The recommendation algorithm based on content filtering has a more serious problem,that is,the new user problem,because the algorithm does not consider the impact of the user's interest change on the recommendation effect.When a new user is added to the system,the new user's browsing history does not exist,and it will not be able to make the right recommendations for new users.In order to solve these problems,this paper proposes a hybrid recommendation algorithm(UIDNN),which combines the user's interest and the nearest neighbor of the domain,and it is used for personalized service recommendation.First,consider the user's interest is not always the same.The change of user's interest with time is similar to that of human being.The nonlinear step wise forgetting function is used to obtain the user's interest in the commodity item.Then according to the user commodity attribute set user interest degree set form,the user and item score collection items not evaluated by using the average value of evaluation items for filling,has formed the user interest degree matrix complementary,reducing the sparsity of data.Secondly,we introduce the nearest neighbor method to find the nearest neighbor of the target user,and reduce the computational complexity of the algorithm according to the degree of user interest.This approach is to determine whether the target user's neighbors have the ability to recommend,so as not to consider those who do not recommend the target users.Prediction of the score of the goods is not evaluated,using the cosine similarity of the user interest degree set to calculate the similarity of the user,and finally to the target user similarity in the size of the first K items recommended to the target user.Based on these recommendations to target users.Through the experiment,the proposed algorithm can be recommended to the userinterest and mixed domain nearest neighbor(UIDNN),based on the similarity calculation method for similarity with Pearson(Pearson),cosine similarity(COS)of two kinds of traditional collaborative filtering recommendation algorithm based on user MAE value comparison according to the experimental results,we can see that the proposed hybrid recommendation algorithm and user interest based on domain nearest neighbor(UIDNN)has smaller mean absolute The smaller the error(MAE),indicating that the UIDNN algorithm proposed in this paper has higher recommendation quality.