Application And Research On Personalized Recommendation Multi-dimensional Aggregation Algorithm

Traditional user-based collaborative filtering recommend algorithm assumes that one score is given by each user for every product,and achieves rating similarity by calculation the Euclidean distance or cosine similarity,which ignores user preferences and the relationship among product attribute,thus the recommendation accuracy is affected.To this end,a metric learning based multi-dimensional aggregation recommend algorithm is proposed,and then this algorithm is applied to hotel recommend.The work of this thesis mainly includes followings aspects:1.Analyze the effect of product dimensional score on the generation of user's nearestneighbor set.And propose to add the user aspect score of product to the user's neighbor setcalculation process,to consider the personalized feature of users.And for missing score,analyze the sentiment orientation of user by sentiment analysis,get the sentiment valueand utilize the value to fill in missing data.Experiment result demonstrates that theexpansion of rating dimension improves the accuracy of recommendation.2.For the problem of product dimension extension in collaborative filtering algorithm,ametric learning based multi-dimensional aggregation algorithm is proposed.Aim to therelational among product attribute,which utilize the Mahalanobis distance to learn thescore matrix of user in same attribution about different products,to complete the acquire ofinitial user-product multi-dimensional score,based these,introduce the aggregationfunction to each user to train regression function.The algorithm constructs the relationshipaggregation function by user's total score and dimensions scores and applies theaggregation function to the initial multi-dimensional score that calculate by metric learning,achieves ratings prediction.The experiment result shows that compared with the previouscollaborative filtering based recommendation algorithm,it has higher recommendationaccuracy.3.According to the significance of hotel recommendation,apply the algorithm proposed inthis thesis to recommend hotels.Then the TOP-K hotels and TOP-N neighbors withsimilarity of target user are generated,it is bring maximum benefit recommendation foruser.