Research Of Personalized Recommendation System Based On LBSN Data

As the wide application of intelligent terminal devices and the rapid expansion of diverse high-precision positioning technologies,location-based social network(LBSN)undergoes a fast development.In the process of development,massive data are generated in LBSN at every second,which leads to data overload issue.To resolve this issue,personalized recommendation algorithm based on LBSN data comes into being.By connecting the physical world with the virtual network world,the LBSN-based personalized recommendation system constructs a check-in preference model for users after analyzing the multi-dimensional data generated by users in LBSN,thereby facilitating users to find the places that meet their specific preferences quickly and accurately from a wide range of choices.Such recommendation system not only saves users' time and energy,but also helps businesses to identify their potential customers and bring advertisement directly to those potential customers,so as to improve business income.With an aim to recommend more accurate services to users,this paper proposes a collaborative filtering algorithm model that combines context information and user preferences and an improved matrix decomposition model,and then embedded these two models into the elderly location recommendation system.This paper conducts the study mainly for several purposes as follows:1.With the rapid development of LBSN,a large number of different types of data are generated in LBSN,although the amount of data is large,but the data is very sparse.In order to solve the problem of sparse check-in data in LBSN-based personalized recommendation system,this paper firstly conducts an analysis of the multi-dimensional heterogeneous data in LBSN to mine the context information of check-in data which may influence user's checkin behavior,and then constructs different models for different context information separately.Finally,a hybrid recommendation algorithm combining context information and user preferences is proposed.2.In the field of recommendation,scholars are mainly put their research focus on the processing of explicit feedback data and have made fruitful achievements by applying the traditional matrix decomposition technology.Different from the previous studies,this paper rolls out research based on implicit feedback data.Although the traditional weighted matrix decomposition model can solve the recommendation problem of implicit feedback data,such model features low solving speed but high time complexity,which cannot meet the real-time and efficient recommendation requirements.Besides,it does not give consideration to the influence of LBSN data diversity on users' check-in behavior.Therefore,this paper proposes a matrix decomposition model that integrates context information and reduces the complexity of the model by K times by using element level least squares alternation method.3.On the basis of the open data set of Foursquare,this paper designs a series of experiments to validate the two improved algorithms and makes a comparative analysis with the previous ones.The experimental outcomes show that the improved algorithms greatly enhance the accuracy and efficiency of recommendation results.4.In order to validate the effectiveness of the above-mentioned improved algorithm in the actual system,and to enable the elderly to quickly find places they may have interests to engage in social and public activities,this paper designs an elderly location recommendation system that has incorporated the improved algorithm.In the end,a sample survey is conducted,which proves that the location recommendation system can indeed help the elderly find locations they are interested in.