Research And Implementation Of Matrix Factorization Recommendation Algorithm Based On Multiple Factors

In recent years,with the popularity of smart mobile phones and the popularity of location technology,it is becoming more and more easy for people to obtain real-time location information in life.This phenomenon has spawned many software applications based on location-based social networks,such as Foursquare and Dianping.The traditional recommendation algorithm has been unable to apply to the user recommendation requirements in the location social network scenario.Different from the traditional recommendation algorithm,there is a large amount of context information in the location-based social network,such as geographic location information,social network information,time information of user behavior,etc.The behavior of the user is the result of the interaction of his own interest preference and various context information.How to properly model these context information in recommender systems becomes a problem that must be considered.The main research contents of this paper mainly include the following parts:(1)This paper proposes a matrix factorization recommendation algorithm based on local social relationship regularization.This paper models the user's social network from the perspective of global social perspective and local social perspective,which makes full use of the influence of user's social network on user preferences.Experiments on real data sets show that the accuracy of the proposed algorithm is better than other social-based recommendation algorithms.(2)This paper proposes a matrix factorization recommendation algorithm based on kernel density estimation.The algorithm considers the impact on user behavior from three aspects:user's own preference,geographic location information and social network information.The algorithm proposed in this paper can not only effectively model the impact of context information on user behavior,but also has good scalability.Experiments on real datasets prove that the proposed algorithm in this paper can improve the quality of recommendation result than other recommendation algorithms.(3)This paper proposes a matrix decomposition recommendation algorithm that fuses temporal effects.Firstly,the algorithm uses the time decay function to weight the user's historical behavior data,then models the user time feature vector and the item time feature vector separately.Finally the model considers the user's recommendation result is based on the user's preference on the item and the user's match with the item.(4)Based on the three innovative recommendation algorithms proposed in this paper,the location recommender prototype system is designed and implemented.The recommender system designed in this paper can combine the results of a variety of personalized recommendation algorithms to users,and can can set the user's personalized algorithm weights.