自注意力下时空-语义相融合的POI序列推荐

POI Sequence Recommendation Model Based on the Integration of Spatiotemporal and Semantics Under Self-attention

近年来,随着基于位置的社交网络(Location-Based Social Network, LBSN)不断发展,POI序列推荐逐渐成为近年来研究的热点问题.现有的POI序列推荐方法仅仅按照时间的先后顺序建模用户历史签到序列,默认用户POI轨迹中连续POI之间具有相等的时间间隔,忽略了用户签到记录之间的时间间隔影响.另外,POI之间的地理距离以及语义信息也是影响推荐准确性的重要因素.基于此,本文提出自注意力下时空-语义相融合的POI序列推荐模型(POI sequence recommendation model based on the integration of spatiotemporal and semantics under self-attention, SA-TDS-PRec).首先,根据用户的实际签到时间建模POI轨迹.其次,融合POI绝对位置、时空间隔以及语义相关信息.最后利用自注意力机制捕捉用户动态偏好的演化,从而提高POI推荐的准确性.在公开数据集Gowalla和Yelp上进行可扩展实验.结果表明,该模型优于目前主流的基准模型,有效提升推荐结果准确性.

As Location-Based Social Network(LBSN) services become increasingly popular, POI sequence recommendation gradually become a hot issue in recent years.The existing POI sequence recommendation methods only model the user′s historical check-in sequence according to the time sequence.By default, there are equal time intervals between consecutive POIs in the user′s POI trajectory, ignoring the influence of the time interval between user check-in records.In addition, the geographical distance and semantic information between POIsare also important.Based on this, this article proposes POI sequence recommendation model based on the integration of spatiotemporal and semantics under self-attention(SA-TDS-PRec).Firstly, to model the user check-in sequence based on actual timestamps;Sequentially, to integrate the absolute position, spatiotemporal interval and semantic related information of POIs.Finally, the self-attention mechanism is used to capture the evolution of users′ dynamic preferences, thereby improving the accuracy of POI recommendations.Extensible experiments on the public datasetsGowalla and Yelp, the experimental results demonstrate theproposed SA-TDS-PRec model to lead to better performance over the state-of-the-art methods.