基于用户轨迹数据的移动推荐系统研究

近年来，随着移动智能设备的普及，移动社交网络方兴未艾，用户习惯和朋友分享自己的精彩经历，因此产生了大规模具有时空属性的用户轨迹数据.从狭义的角度来看，轨迹数据是指连续采样的GPS数据.从广义的角度来看，在时空域存在连续性的序列，都可以称作轨迹.例如：在社交网络上的用户签到序列就可以认为是粗粒度的轨迹数据.广义轨迹数据具有时空异构性、连续与离散并存、时空项目的层次性不明显和分类不明确等特点，但是相比于GPS轨迹数据，广义轨迹数据来源广泛，蕴含丰富的信息，这给传统的移动推荐系统带来了巨大的机遇.与此同时，广义轨迹数据规模大、结构丰富，这也给传统的移动推荐系统带来了巨大的挑战.如何利用广义用户轨迹数据来提升移动推荐系统的性能，已成为学术界和产业界共同关注的重要课题.以轨迹数据特征作为切入点，对近年来基于广义用户轨迹数据的移动推荐系统的主要模型方法和推荐评价指标进行了系统综述，阐述了与传统移动推荐系统的联系和区别.最后，对基于广义用户轨迹数据的移动推荐系统有待深入研究的难点和发展趋势进行了分析和展望.

Survey on Mobile Recommender Systems Based on User Trajectory Data

In recent years, with the popularity of mobile smart devices, location based social networks are on the rise. Users trend to share their wonderful experiences with their friends, resulting in producing large-scale user trajectory with temporal and spatial attributes. From a narrow perspective, the trajectory data refers to continuously sampled GPS data only. From a broad perspective, it can be called trajectory data as long as the data has sequential characteristic. Thus, the check-ins, acquired from a social network, can also be considered coarse-grained trajectory data. The generalized trajectory data has the characteristics of spatiotemporal heterogeneity, continuous and discrete coexistence, and containing temporal-spatial items with unclear hierarchy and classification. However, compared to the GPS trajectory data, the generalized trajectory data source is extensive and contains rich information, which brings great opportunity to the traditional mobile recommender system. At the same time, the generalized trajectory data has big scale and diversity structure, which also presents great challenges to the system. It has become an important issue how to use the generalized trajectory data to improve the performance of mobile recommender system in academia and industry. This paper takes the trajectory data characteristics as the focal point to analyze and survey main recommender methods and evaluation metrics based on generalized user trajectory data. Further, it expounds the relationships and differences between traditional mobile recommender systems and the mobile recommender systems based on user trajectory data. Finally, the paper discusses the difficulty and development trend of mobile recommender systems based on generalized user trajectory.