社会化媒体中的移动轨迹语义增强和挖掘推荐研究

Effectively managing massive travel trajectories in social media and discovering interesting patterns from them are of great importance to answering various advanced user queries in location-based services. However, multimedia content and social semantics associated with trajectories are not well exploited in traditional trajectory data mining. This project will investigate the basic problem of how to fuse visual, tag and location information in social media. Since they are correlated, we will study multimodal data fusion and reconstruction, as well as trajectory semantic enhancement for multilayered pattern mining and analytics, in order to finally overcome the obstacles in trajectory data mining such as trajectory uncertainty and data sparseness. Moreover, through user interest and query intention understanding, we will develop a system of personalized route recommendation and visualization for travel applications. The project will produce a series of key theories and technologies for synergy of social media and mobile context, which will lay the foundation of applications such as semantic understanding and mobility analytics, and promote the applicable range of media big data in relevant disciplines.

对社会化媒体中的海量移动轨迹进行有效地管理，并从中发现有意义的模式来满足位置服务中用户多样的查询需求是一个重要的研究问题。传统轨迹数据挖掘未能充分考虑与轨迹相关的多媒体内容和社会化语义。本项目将探索社会化媒体中的视觉、标注和位置信息相互融合的关键基础问题，从三者的关联分析出发，研究多数据形态融合重构以及轨迹语义增强，实现移动轨迹的多层次模式挖掘与分析，最终突破轨迹数据挖掘中的不确定性和稀疏性等瓶颈。在此基础上，进一步研究用户兴趣与查询意图理解，并结合导航和旅游等应用，构建路径个性化、可视化的推荐系统。本项目拟输出一系列社会化媒体和位置情境协同分析的计算理论和技术体系，为语义理解、移动分析等应用奠定基础，并促进媒体大数据在跨相关学科中的应用。

本项目的研究关注协同利用多媒体内容和社会化上下文与移动轨迹的方法，以视觉-标注-位置融合的统一表达为线索贯穿轨迹数据的语义增强、挖掘和可视化推荐等多个环节。初步建立了基于数据关联的知识推理与语义增强方法、探索了对轨迹数据确定性-不确定性统一理解的框架、提出了面向多种社会化媒体数据的融合多角度用户模型，开发了文本和图像/视频等与轨迹结合的优化展示技术。目前共计发表论文49篇，获得授权发明专利5项，依托项目培养了20余名研究生，2019年主办了CCF推荐国际会议APWeb-WAIM。部分所研发的轨迹分析和视频监控关键技术在四川九洲电器集团有限责任公司的相关实际系统中得到实际应用。