The Research And Implementation On Restaurant Recommendation Model Based On Multi-source Information And Location Migration

Restaurant recommendation can leverage check-ins,time,location,restaurant attributes and user demographics to dig user,s dining preference,and recommend a list of restaurants for each user.However,most of the current research still faces the following challenges:1)the difficult to fuse these data information,2)cold start problem,including new user and new item,3)out-of-town recommendation,when users travel out of town,the traditional collaborative filtering is difficult to effectively recommend.In light of the above,the contributions of this paper are summarized as follows.(1)In order to fuse these data information more effectively,we propose a restaurant recommendation model with multiple information fusion(RRMIF).Firstly this model constructs a three-dimensional tensor,some users' similar relation matrices and restaurants' similar relation matrices from user's check-ins and additional data information.Secondly these relation matrices and tensor are decomposed simultaneously.Then bayesian personalized ranking optimization criterion method(BPR Opt)and gradient descent algorithm are adopted to solve the model parameters.Finally our model generates a corresponding restaurant candidate list for target user at different time by calculating predicted tensor.(2)In order to alleviate the cold-start problem and model user's decision,most of the current research is to integrate social network information,and to mine user preferences through friend preferences,but new users and new items are still not fully learned.Therefore we propose a restaurant recommendation model based on user activity area and social network(UAASN).The model first defines three types of restaurants based on user activity area and check-ins,including signed-in restaurants,potential check-in restaurants,and out-of-area restaurants.At the same time,the model considers friend preferences and expert preferences to influence user preferences,and defines the surrounding area of the check-in restaurant as the user activity sub-area to improve the possibility of signing in the surrounding restaurants.Finally construct the error function by using two-layer bayesian personalized ranking optimization criterion,and adopt the gradient descent algorithm to solve the model parameters.(3)In order to overcome out-of-town recommendation,a restaurant recommendation model based on user location migration is proposed(RRULM).The model constructs collaboration preference based on the restaurant category and social friends,the topic preference based on the content information,the word-of-mouth preference based on the regional cluster.Then Logistic regression is used to mixes the above preferences,the model improves the accuracy of out-of-town recommendations to some extent.(4)Based on the proposed model,a restaurant recommendation system based on multi-source information and location migration is designed and implemented.The system has certain practicability,which combines different strategies for different scenarios to recommend,and further utilizes user explicit feedback to optimize the ranking.