跨语言社会舆情分析基础理论与关键技术研究

跨语言社会舆情分析是智能信息处理的一个研究热点。本项目旨在建立一套适用于跨语言社会舆情分析的理论方法体系，开发一系列关键技术，并构建相关资源和实验平台，解决阻碍其发展的关键科学问题。特别是，1)针对社会舆情的篇章属性，基于篇章结构和话题结构理论，开展篇章级事件信息抽取研究；2）针对社会舆情的难预测性、突发性与多变性，基于事件关系网络，实现社会舆情的准确预测、及时发现与快速跟踪；3）针对社会舆情的多源异构性，基于贝叶斯网络，开展面向社会舆情的事件可信度计算研究；4）针对社会舆情传播的跨语言特性和不同语言资源的不对称性，探索跨语言推理机制、资源迁移和语义空间等跨语言技术，实现面向社会舆情分析的跨语言信息处理；5）针对社会群体情感的多面性，基于社会关系网络，开展社会群体情感分析研究；6）针对资源匮乏问题，基于互联网大规模自然标注资源、自举学习和群体智慧，构建相关资源，保障本项目的顺利开展。

Cross-lingual social opinion analysis has been an important and hot research topic in recent years. However, there exist various kinds of problems in current studies, such as passivity, non-objectivity, and lack of deep understanding and analysis. This project attempts to establish a theoretical infrastructure suitable for cross-lingual social opinion analysis, develop a series of key techniques, construct relevant resources, and build a cross-lingual social opinion analysis platform, in addressing various scientific obstacles in front of its wide application. Especially, .1) To confront the discourse characteristics of social opinion, this project proposes a discource structure and topic structure-driven framework to carry out discourse-level event information extraction..2)To confront the prediction difficulty, sudden occurrence and fast change of social opinion, this project proposes an event network-driven framework to achieve accurate prediction, timely identification and fast tracking..3) To confront the multi-source and different structure characteristics of social opinion, this project proposes a Bayes network-driven framework to carry out reliability computation of its content..4) To confront the cross-lingual characteristic of social opinion broadcasting and the asymmetric of different language resources, this project proposes a series of cross-lingual techniques, such as entity linking, document inference and semantic space, to achieve social opinion analysis-driven cross-lingual information processing. .5) To confront the opinion variability of different social groups, this project proposes a social network-driven framework to achieve social group-based opinion analysis..6) To confront the lack of language recourses, this project explores large-scale naturally-annotated web resources, bootstrapping and collective intelligence to construct relevant resources in supporting smooth carry-out of this project.