Research On Intelligent Recommendation Algorithm Based On User Interest Concept Lattice

Recommender system can select valuable information from a lot of data and help users find the products or services they need.It has been widely used in recent years.However,there are still many problems in the current recommendation system,such as sparse data,natural noise and cold start,which leads to the failure of the recommendation system to obtain accurate user preferences.This thesis mainly focuses on the research of natural noise and data sparsity.In the recommender system,when users score products,the data of recommender system will be inconsistent.These inconsistencies,the so-called natural noise,will affect the recommendation results.In order to correct the natural noise in recommendation system,this thesis proposes a collaborative filtering correcting natural noise based on concept lattice(CFCNN-CL)algorithm.Firstly,users and items in the scoring matrix are divided into three categories: strong,average and weak.Natural noise is detected by analyzing the contradiction between categories.Then,collaborative filtering based on concept lattice is used to predict new scoring values to correct these noise ratings.Finally,unrated items are predicted from the data set without natural noise.In the actual recommender system,the scoring data provided by users is very rare.When users only score a few items,the quality of recommendation will be greatly affected.In order to solve the problem of poor recommendation effect of existing recommendation algorithms in sparse data sets,this thesis proposes a recommendation rating prediction based on user interest concept lattice(RRP-UICL)method.Firstly,the nearest neighbor is divided into direct nearest neighbor and indirect nearest neighbor by user interest concept lattice.Then different methods are used to calculate the similarity between direct "nearest neighbor" and indirect "nearest neighbor" and target users.Finally,the invisible item score of the target user is calculated by the similarity value.Experimental results on real datasets show that the proposed CFCNN-CL algorithm and RRP-UICL algorithm have high recommendation accuracy,and still have good performance in the case of sparse data.