基于物理空间与网络空间的社会事件分析的研究

In recent years, the fast development on internet and digital technology and greater focus on social events from the governments and society have made the research on social event analysis based on information from cyber and physical spaces (bi-space) on an urgent need. Recent studies mostly focused on analysis from a single separated space or media while neglected the correlation between the cyber and physical spaces. This project targets to establish a real-time, efficient and robust social event analysis model which can analyze the correlation on information from both cyber and physical spaces on a provided social event. The model can mine each sub-event occuring different times and locations, and provide evidences from real-world physical space according to the topics in cyber space, as well as the confidence for samples in cyber space. To achieve that, the project will firstly integrate the study on fusing the results of natual language processing and image processing to analysize and understand the semantic information from the cyber and physical spaces on the provided event to construct "words-timestamp-location" structured feature; and then develop technology on semantic analysis across-spaces, cross-media and cross-spatio-temporal to reveal correlations among the different datasets from two spaces; lastly, using this system to design a general structured cross-modality co-clustering framework, and utilize correlations among clusters and the source of each sample to mine the clusters with the same topic, as well as the confidence score for each sample from cyber space.

近年来，随着照相设备的飞速发展、互联网技术的全面应用以及国家和社会对社会事件的广泛关注，使得基于物理空间和网络空间（二元空间）的社会事件分析的研究势在必行。现有研究大都局限在某个空间或某个独立媒体库上的社会事件，而忽视了二元空间的相互关系。本课题旨在研究基于二元空间的社会事件的分析技术，建立实时、有效、鲁棒的社会事件分析模型，对给定的社会事件，挖掘其各阶段在不同时间地点发生的子事件，提供二元空间中与之相关的样本子集，并给出网络空间样本的置信度。基于此，本项目将首先对二元空间上与给定的社会事件相关的信息样本进行基于"词-时间-地点"结构的语义分析；继而分别对多源样本在跨空间、跨媒体、跨时空三个层次上实现语义上的关联，根据"时间-地点"信息挖掘社会事件的各个子事件；最终，建立一般化结构的多模态样本协同聚类框架，发现属于同一子事件的样本聚类，在物理空间和网络空间上，建立样本置信度的传播模型。

随着互联网、物联网技术的全面应用，以及国家、人民对社会热点事件的广泛关注，使得在物理空间和网络空间下的数据感知和社会事件检测分析的研究势在必行。本课题旨在研究物理空间与网络空间的社会事件分析的关键技术和科学问题，包括二元空间的信息理解、二元空间复杂关系的发现和二元空间的协同分析。构造以热点事件数据集为主的核心数据，提出网络空间信息理解、物理空间信息理解、跨媒体语义关联、跨时空语义关联、跨二元空间语义关联、基于协同聚类的二元空间事件检测等技术。可以为国家和地区的公共安全检测提供一整套科学技术支持。