Research On Recommendation Algorithm Of Restaurant Based On User Trust And Location Preference

With the Internet stepping into Web 2.0 era,O2 O e-commerce rapidly rises and infiltrates into all aspects of people's life services.While bringing convenience to users,it also creates profits for merchants.The restaurant field is an important part of it.However,with the increasing number of users and businesses on the website,e-commerce website has experienced a serious information overload problem,so the personalized recommendation system has been widely used as an effective technique.This thesis studied the personalized recommendation of the restaurant field in O2 O e-commerce,the research points are listed as follows:1.Aiming at the problem of user-restaurant rating data sparsity in traditional restaurant recommendation system,a collaborative filtering recommendation algorithm based on conditional probability filling pf_UCF is proposed.Firstly,pf_UCF algorithm selects all restaurants whose average rating are same as the restaurant to be evaluated from the user's rated restaurants,and calculates the statistical probability that the user rates 1~5 points for these restaurants.Then take the rating corresponding to the maximum probability value as the user's rating for the restaurant,and fill the rating into the userrestaurant rating matrix.Finally,based on the filled rating matrix,a collaborative filtering algorithm(CF)is used to generate recommendation result.Experimental results show that the pf_UCF algorithm can effectively alleviate data sparsity and improve the accuracy of restaurant recommendation.2.Aiming at the problem that the traditional CF restaurant recommendation algorithm does not consider that the dining choice of users is affected by the context factor,a restaurant recommendation algorithm based on the user's location preference ULPRR is proposed.Firstly,ULPRR algorithm establishes a location feature vector for each user based on the user's historical rating data,and uses the vector to calculate the user's location preference similarity.Then,the user's location preference similarity,interest similarity and the rating similarity obtained by partially filling the user's rating vector are combined,and the final user similarity is applied to the CF algorithm for recommendation.Experimental results show that the ULPRR algorithm can effectively improve the effect of restaurant recommendation.3.Aiming at the problem that the traditional CF restaurant recommendation algorithm does not consider that the dining choice of users is affected by their potential trusted friends,a restaurant recommendation algorithm based on user trust relationship UTRR is proposed.Firstly,UTRR algorithm builds a global trust for each user,and calculates trust degree between users with explicit trust relationship,mines implicit trust relationship for users without explicit trust relationship.Then,based on the improved trust matrix,the trust-based CF is used to predict user's rating for restaurant.Finally,combine this rating with the one predicted by bias singular value decomposition to produce the final recommendation result.Experimental results show that compared with the traditional trust-based recommendation algorithm,the UTRR algorithm has obvious superiority and improves the quality of restaurant recommendation.