面向图文混合的网络舆情新事件发现及其关联挖掘

Currently, the co-occurrence data that combines text and image is very popular, and has become the mainstream of event propagation. Moreover, with the help of the efficient broadcasting of social media with effective expression of such data, it has received more concentration on the importance and effectiveness of event propagation in the Internet. By means of event correlation in public opinion space, it leads to boosting and accelerating of the evolution of public opinion. Here we focus on the new event detection, correlation mining and tracking in co-occurrence environments of texts and images based on some machine learning methods, such as Deep learning and Extreme Learning Machine. Specifically, we focus on resolving the text and image data fusion, event correlation modeling, implicit correlation metrics in public opinion space, and some other key issues. In general, we propose a method for new event detection and their correlation mining, and thus provide core technical support for public opinion spatial visualization rendering, event correlation automated building and dynamic tracking. We will publish 10-20 high-level academic papers, including more than 10 SCI/EI journals, and apply for 3-4 patents and 3-4 software copyrights.

目前，图文混合数据已经成为舆情事件传播的主要形态。借助于社交媒体高效的传播机制和图文混合数据有效的表现形式，网络环境下舆情事件的重要性和影响力越来越受到关注。舆情事件之间互相关联，引发，助推并加速舆情的演变和激发新事件。本课题以图文混合环境下舆情新事件发现与事件关联动态跟踪为目标，基于深度学习与极速神经网络等数据挖掘技术，重点解决图文数据深度融合、事件关联多维建模、隐式关联度量体系构建等关键问题，研究一种面向大规模图文混合数据的舆情新事件发现及其关联挖掘方法，为开展舆情空间可视化呈现、事件关联自动构建与动态跟踪、关联状态演化异常监测等舆情服务提供核心技术支持。发表 10-20 篇高水平学术论文，其中 SCI/EI 期刊 10 篇以上； 申请发明专利 3-4 项，软件著作权 3-4 项。

本课题以图文混合环境下舆情新事件发现与事件关联动态跟踪为目标，基于深度学习与极速神经网络等数据挖掘技术，重点解决了图文数据深度融合、事件关联多维建模、隐式关联度量体系构建等关键问题，研究了一种面向大规模图文混合数据的舆情新事件发现及其关联挖掘方法，为开展舆情空间可视化呈现、事件关联自动构建与动态跟踪、关联状态演化异常监测等舆情服务提供了核心技术支持。.在项目实施过程中，形成了一系列论文、专利、原型系统等成果。在IEEE Transactions on Neural Networks and Learning Systems、Neural Networks、IEEE Intelligent Systems、Neurocomputing等国际知名期刊发表论文20余篇，获得陕西省高等学校科学技术二等奖，申请发明专利5项，软件著作权3项。