关键词: 下一个兴趣点推荐, 时序行为, 注意力机制, 社交关系

Abstract: Next point-of-interest (POI)recommendation is one of the key applications in geolocation-based social networks.To address the issues of inadequate representation of users' cyclic temporal behavior and insufficient mining of social relationships in existing methods, this paper proposes a POI recommendation method that integrates both cyclic temporal behavior and social relationships.We analyze users' behavioral patterns from their check-in sequences across three time granularities: short-term, periodic, and long-term, and extract cyclical time-sequential behavioral features.Additionally, we mine social relationships between users by examining the overlap in their check-in records and the similarity of their friends, extracting dual-layer social features.The method introduces feature fusion with an adaptive weight allocation strategy and calculates users' preference scores for POIs.Based on these scores, the next POI is recommended to the user.Experimental results on the Sina Weibo (Shanghai)and Foursquare (New York)datasets demonstrate that the proposed method significantly improves hit rate (HR)and normalized discounted cumulative gain (NDCG).

Key words: next point of interest recommendation, sequential behavior, attention mechanism, social relationships