Research On Point Of Interest Recommendation Of Mobile User Group Based On Location Based Social Networks

With the rapid development of big data,the recommendation system in the new situation should be built,and everyone is going to be a "transparent person".The information on the Internet includes shopping,music,knowledge,social behavior,status and location that have the significant value to be used.Everyone will be the information provider,also the information user.The central of the recommendation system is to provide commodity information and decision making which will be in accordance with personalized demand of different users.As a Recommendation system,it is to analyze interest preference and social behavior,and provide information by these characters.With the prominence of the E-commerce,the quantity and variety of goods growing rapidly,and individuals need to spend a lot of time on finding the goods which can satisfy their demand.It is wasteful to spend large amounts time on looking for irrelevant information which will cause information overload problems.In the mobile social networks,the Location Based Social Networks(LBSNs)supports the records of geographical and check-in information of the user,gets the user's behavior path,and provides more personalized and convenient services for the users.Therefore the LBSNs has become a hot issue in the theoretical and practical.In the existing literature research,it is found that the state-of-the-art recommendation algorithm is not suitable for the mobile users with frequent location changes,and the calculation of user similarity is not accurate enough.But with the increasing number of users in mobile commerce,it is particularly important to increase the accuracy of recommendation and reduce the time for users to gather information.In this paper,three large real-world check-in dataset(Foursquare LBSN dataset,Gowalla LBSN dataset,Brightkite LBSN dataset)are used to access the data,and the Mobile Relationship and Spatio-Temporal Model(MRST)is proposed.The Improved Kernel Density Estimation,Relative Entropy and KL divergence are used,and describe the user mobility and randomness by Information Entropy,and express user similarity by Jaccard coefficients.In addition,the changes of user similarity by the strength of Relationship are considered and provide the definition of strong and weak relationship.At the same time,both time and geographical factor which make effect of user similarity are considered.Therefore,this paper integrates user interest,user similarity,time and geographic factor organically to achieve a recommendation approach.Moreover,the number of check-in recorded according to time factor is small,there will be some missing mobile information in the process of collecting data,which resulting in incompleteness of information.Therefore,it is an effective approach to alleviate data sparsity and cold start problem through mobile recommendation by circle of friends,and effectively improve the accuracy and practicability of recommendation results.The experimental results show that,compared with the traditional recommendation,the model proposed in this paper can effectively predict the user's next POI,get the best list of user recommendation,and improve the accuracy and recall rate significantly.The results will provide strong technical support and decision-making services for how to better satisfy the user personalized needs and precise recommendation in mobile commerce.