| With the promotion of Internet technology and intelligent communication equipment in recent years,network applications have been popularized on a large scale.How to accurately provide users with personalized services has become the focus of academic and industrial circles,which has further given rise to a research boom on recommendation technology.Location-based social networks have become an information exchange platform for a wide range of audiences,so it is of great theoretical significance and commercial value to study the next point-of-interest recommendation method.The results of recommendation tasks are largely subject to user preferences.However,under the background of massive data and diversified scenarios,the characteristics of variability and complexity of user preferences make it so difficult to learn user preferences,thus affecting the performance of recommendation.At present,the research on recommendation of the next point of interest mainly faces the following three challenges: how to mine user interest in the context of sparse and complex data,how to learn the impact of different preferences on future checkin,how to extract the cooperation information between users and interest points for personalized recommendation.Combined with the above analysis,the main research contents of this paper are as follows:In view of the complex and changeable characteristics of user behaviors that hard to capture,a network framework integrating attention mechanism and adaptive convolution driving is designed to improve the recommendation performance of the next point-of-interest.In order to accurately depict the user behavioral preferences,the long-and short-term memory network is used to extract the context feature information from the historical checkin,and on this basis,the multi-head self-attention mechanism is introduced to capture the user long-term preferences.For the dynamic interest preference contained in the short-term behavior of users,an adaptive convolutional network is constructed,which selectively aggregates multiple parallel convolution kernels and autonomously regulates the learning of personalized preference according to the check-in data.Finally,when combining user longterm and short-term preferences,different preferences contribute differently to user next decision making.The limitations caused by simple fusion are avoided by giving weight to the two preferences through the fusion mechanism.Aiming at the problem that it is difficult to fully extract the cooperative information between users and point-of-interests,a framework of dynamic association graph enhanced network is proposed,which establishes the association graph from the perspective of users and point-of-interests respectively to obtain the information changing over time.More specifically,an associative information perception unit is designed to enhance the dynamic representation ability of nodes with hope to learn the interaction process between users and point-of-interests across time,category and other feature information by extending graph convolutional neural network.Then,through stacking the unit to capture the higher-order connectivity information between the user and the point-of-interests,the next interest pointof-interest recommendation can be more effective.Based on the public data sets,the two proposed next point-of-interest recommendation methods are experimentally verified and compared with the existing baseline methods.The results show that the proposed methods can mine users interests better and capture the interaction between users and interest points more adequately than the baseline method. |
