高性能计算环境下社交媒体地理大数据热点挖掘与智能推荐

In the era of big data, the increasing popularity of the applications which share the social media location has led to the producing of large amounts of geotagged data. It is the cutting-edge research that how to extract abundant individual behaviors and characteristics from the geotagged big data. Our research, from the perspective of multi-modal massive data environment and high-performance computing environment, combines the knowledge of geography with the theory and approach of machine learning, complex network and space-time computing. We expect to construct a high-performance hotspot mining framework for geotagged social media data, to provide the deep perception of the place, route and network from various perspectives, to explore the space-time distribution pattern of geographical hotspot. We then propose an intelligent recommendation model for individual travel, providing accurate service and decision support for public travel. In general, our research mainly focuses on building the theory of cognition and representation, processing approach and analytical method of the hotspot, based on the fourth paradigm, which includes place, route and network. Our research is theoretically meaningful. It may reveal the unknown knowledge and laws of geography, breaking the traditional research system of induction, deduction and simulation as the main features. In addition, the research of intelligent recommendation model can help to realize the "smart travel" of residents, and promote the development of related industries such as catering, accommodation, entertainment and tourism industries. It, therefore, has a broad vision of application.

大数据时代下，社交媒体用户分享位置信息的应用日益普及，由此产生了海量、形式多样的社交媒体地理数据。如何挖掘其中蕴含丰富的个体行为与用户特征，已迅速成为业界前沿。本项研究拟在海量多模态新数据环境与高性能新计算环境综合视角下，融合地理领域知识与机器学习、复杂网络与时空计算等理论与方法，构建面向社交媒体地理大数据的高性能热点挖掘框架，实现对场所、线路与网络的多角度深层感知，探索地理热点的时空分布模式与规律，并建立面向个体出行的智能推荐模型，为居民的智慧出行的提供精准服务与决策支持。综合而言，本课题期望建立基于第四范式的地理热点（含场所、线路与网络）认知与表达理论、处理与分析方法，突破传统的以归纳、推演与仿真模拟为主要特征的研究体系，发现过去所未知的地理知识及规律，具有重要的理论价值。此外，智能推荐模型的研究可帮助实现居民的“智慧出行”，带动“食住娱游”等关联产业的发展，具有广阔的应用前景。

大数据时代下，社交媒体用户分享位置信息的应用日益普及，由此产生了海量、形式多样的社交媒体地理数据。如何挖掘其中蕴含丰富的个体行为与用户特征，已迅速成为业界前沿。本项研究在海量多模态新数据环境与高性能新计算环境综合视角下，融合地理领域知识与机器学习、复杂网络与时空计算等理论与方法，构建面向社交媒体地理大数据的高性能热点挖掘框架，实现对场所、线路与网络的多角度深层感知，探索地理热点的时空分布模式与规律。具体来说，通过本项研究主要在以下三个方面取得了重要进展: 1. 研究海量社交媒体地理数据的NoSQL存储组织方法；2. 构建高性能地理热点挖掘方法体系，提出基于自适应空间聚类的场所发现方法、基于序列模式挖掘的线路发现方法与基于邻接场所的有向加权复杂网络模型；3. 建立基于随机森林与隐语义分析的级联式智能出行推荐模型。.依托本研究共计发表论文19篇，其中SCI/SSCI论文19篇。发表期刊包括遥感与地理信息科学主要刊物JAG, IJGIS, Transactions in GIS, CEUS等,以及地理学旗舰刊物AAAG。上述成果取得了国内外同行的认可，产生了较高的学术影响。本项目的基于用户个性化的线路优化系统，已经用于深圳市坪山区的“新公交”项目，首期工程已经落实到位产生了良好的社会效益。.