深度学习方法在兴趣点推荐中的应用研究综述

在基于位置的社交网络（LBSN）中，用户可以在兴趣点（POI）进行签到以记录行程，也可以与其他用户分享自身的感受并形成社交好友关系。POI推荐是LBSN提供的一项重要服务，其可以帮助用户快速发现感兴趣的POI，也有利于POI提供商更全面地了解用户偏好，并有针对性地提高服务质量。POI推荐主要基于对用户历史签到数据以及用户生成内容、社交关系等信息的分析来实现。系统归纳POI推荐中所面临的时空序列特征提取、内容社交特征提取、多特征整合、数据稀疏性问题处理这4个方面的挑战，分析在POI推荐中使用深度学习方法解决上述问题时存在的优势以及不足。在此基础上，展望未来通过深度学习提高POI推荐效果的研究方向，即通过增量学习加速推荐模型更新、使用迁移学习缓解冷启动问题以及利用强化学习建模用户动态偏好，从而为实现效率更高、用户体验质量更好的推荐系统提供新的思路。

A Survey of Studies on Deep Learning Applications in POI Recommendation

In Location-Based Social Network(LBSN), users conduct check-ins at selected Point of Interest(POI) to record their trajectories, share feelings with their friends and form social friends.POI recommendation is an important service of LBSN to help users find POIs that meet their interests.The service can also help service providers understand user interests and improve user experience accordingly.POI is implemented mainly by analyzing the historical check-in data, social relationships, reviews and other user information that can be explored to speculate user preference.In this paper, we summarizes challenges faced by POI recommendation, i.e., spatial-temporal sequential feature extraction, social feature extraction from user content, multi-feature incorporation, and solutions to data sparsity.Then we give a brief introduction to deep learning-based methods for POI recommendation, and review their advantages as well as disadvantages displayed when dealing with the above challenges.On this basis, we point out future directions of studies on deep learning applications in POI recommendation, i.e., incremental learning to accelerate the recommendation model update process, transfer learning to solve the cold start problem, and reinforcement learning to model dynamic user preference.The discussion attempts to provide reference to studies on improving the efficiency and user experience of recommendation systems.