Studying the POI recommendation in an online setting becomes meaningful because large volumes of user-POI interactions are generated in a chronological order. Although a few online update strategies have been developed, they cannot be applied in POI recommendation directly because they can hardly capture the long-term user preference only by updating the model with the current data. Besides, some latent POI information is ignored because existing update strategies are designed for traditional recommder systems without considering the addtional factors in POIs. In this paper, we propose an Online Meta-learning POI Recommendation (OMPR) method to solve the problem. To consider the geographical influences among POIs, we use a location-based self-attentive encoder to learn the complex user-POI relations. To capture the drift of user preference in online recommendation, we propose a meta-learning based transfer network to capture the knowledge transfer from both historical and current data. We conduct extensive experiments on two real-world datasets and the results show the superiority of our approaches in online POI recommendation.

在在线环境中研究 POI 推荐变得有意义，因为大量的用户-POI 交互是按时间顺序生成的。尽管已经开发了一些在线更新策略，但它们不能直接应用于 POI 推荐，因为它们仅通过使用当前数据更新模型很难捕捉到长期的用户偏好。此外，一些潜在的 POI 信息被忽略，因为现有的更新策略是为传统的推荐系统设计的，没有考虑 POI 中的附加因素。在本文中，我们提出了一种在线元学习 POI 推荐 (OMPR) 方法来解决该问题。为了考虑 POI 之间的地理影响，我们使用基于位置的自注意力编码器来学习复杂的用户-POI 关系。为了捕捉在线推荐中用户偏好的漂移，我们提出了一个基于元学习的迁移网络来捕捉历史和当前数据的知识迁移。我们对两个真实世界的数据集进行了广泛的实验，结果表明我们的方法在在线 POI 推荐方面的优越性。