Research On Business Recommendation Method Based On Constructing Multi-attribute Matrix

With the rapid development of information technology,the information on the network has exploded.How to effectively use the effective information in the data to complete recommendations to users has become a research focus in the network ecology.In the evaluation data of the business review website,the business attributes and user emotions are implied.The existing business personalized recommendations mainly generate recommendations based on the rating information of neighbor users,which can not explore the user preferences at a fine-grained level,ignoring the emotional rating of businesses and users ignoring the emotional ratings of businesses and users at the attribute level.In this regard,this paper proposes a business recommendation method based on building a multi-attribute matrix.Firstly,in view of the problem of insufficient research on attribute fine-grained sentiment analysis at the attribute level of business and users,a multi-attribute business recommendation method MFAMM combining time series and user preferences is proposed.The MFAMM method extracts the attributes of the business,and completes the classification of the attribute class by clustering;performs sentiment analysis on the comments containing attribute category words,and calculates the average value of sentiment to obtain the user's emotional tendency score for the attribute category.And according to the factors that users' preferences will change with time and opinions authority will be different due to different users' experience,time sequence factor and users' opinions weight factor are added to the MFAMM method respectively.The sentiment analysis results and multi-attribute factors are combined to construct the user preference attribute matrix and the business attribute scoring matrix,and use similarity matching method for recommendation.Secondly,for the problem of unstable recommendation effect caused by the sparseness of the matrix,the AMF algorithm is proposed to fill the attribute matrix.On the basis of the Slope One algorithm,the similarity of the business attribute layer is added to improve the accuracy of the recommendation,reflecting the user's interest at the fine-grained level,and alleviating the data sparsity problem of the user-merchant rating matrix.Based on the user rating matrix,the similarity between users and the nearest neighbor of user preference is calculated,which realizes the filling of attribute matrix and the matching of user and business attribute matrix recommend.Finally,for two different matrix matching business recommendation algorithms,experiments are carried out on Yelp public data sets,and the algorithm proposed in this paper is compared with similar recommendation algorithms.It is verified that the MFAMM algorithm has certain advantages in recommendation accuracy and recall rate,and AMF The algorithm effectively improves the sparsity of the matrix,improves the stability of the recommendation effect,and makes the matrix matching recommendation get better results.