Research On Next Point Of Interest Recommendation Method Based On Multi-Level Attention Mechanism And Long-and Short-Term Preference Modeling

With the continuous development of Internet technology and popularization of electronic devices,a large amount of information has entered the stage of global circulation,and the development of social networks is also facing unlimited opportunities and challenges.Among them,the POI(point-of-interest)recommendation problem in location based social network(LBSN),as one of the most important applications,has a wide application prospect.LBSN provides location-based services that enable users to share their location and lives by checking in to places such as Dianping,Meituan,Facebook and Yelp.The main goal of the next POI recommendation is to predict the next POI a user is likely to visit within a given time frame based on the information available in the user’s check-in sequence.The next location recommendation plays an important role in location-based services,not only improving the quality of location-related business services,but also helping to enhance the customer experience.In order to solve the sparse check-in problem of the next POI recommendation in LBSN,improve the accuracy of recommendation and the service satisfaction of users,this paper makes full use of the check-in time,geographical location,POI category and other relevant information in LBSN,and proposes a next point of interest recommendation model(MALS)based on multi-level attention mechanism and long-and short-term preference modeling.MALS considers the user’s category preference,long-term preference and short-term preference respectively,and further studies the impact of these three preferences on the final user preference,which is divided into four modules: 1)The category module uses the combination of LSTM,long-and short-term memory(LSTM)network and temporal attention mechanism to study users’ category preferences at the level of coarse-grained POI semantics(category),and participates in the calculation of long-term and short-term preferences of subsequent users as an auxiliary module;2)The long-term preference module combines LSTM with multi-level attention mechanism to solve the problem of data sparsity by considering multiple context information of user check-in behavior and category preferences of users.In addition,multi-level attention mechanism studies the multi-factor dynamic representation of user check-in behavior and the nonlinear dependence relationship between user check-in.Get users’ long-term preferences at the fine-grained POI level;3)The short-term preference module uses a recurrent neural network(RNN)and temporal attention mechanism to effectively improve the computing efficiency of the model,research the nonlinear dependence between short-term check-in and obtain users’ short-term preferences combined with category preferences;4)The output module generates user preferences by combining the long-term preferences and short-term preferences of users,and designs a filter to filter all POIs through certain filtering rules to obtain each user’s candidate POI.The final POI probability ranking list is formed through the calculation of user preferences and candidate POI.In this paper,two public real world check-in datasets in Foursquare,Charlotte(CHA)and New York(NYC),are used for experimental comparison and comprehensive evaluation of MALS.The results show that MALS is superior to other comparison algorithms in terms of recall rate and mean average precision.