III: Small: Multi-Dimensional Structuring, Summarizing and Mining of Social Media Data

各种各样的社交媒体已经影响了数十亿用户在全球范围内获取和分享信息的方式。这创造了巨大的机遇，但也带来了巨大的挑战，理解，总结和挖掘这些数据，由于其庞大的体积，以及动态和非结构化的文本内容的性质。针对这些挑战，本项目以基于文本的社交媒体为研究对象，提出了一种多维数据结构化方法，通过挖掘非结构化社交媒体数据，揭示其隐藏的多维结构。该项目研究社交媒体结构、汇总和挖掘的原理、方法和算法，并为多维社交媒体数据分析开发有效且可扩展的技术。本研究的原理和方法也可以推广到其他类型的大规模非结构化数据的可扩展和多维分析中。为了进行有效的多维社交媒体结构，该项目开发了一种基于远程监督的方法，将人工管理和标签的工作量降至最低。它以维基百科、Freebase或其他知识库中的数据为参考，将社交媒体数据与相应的新闻或其他相关文档集成，进行短语挖掘、实体和事件发现和键入，并从社交媒体中发现与这些实体和事件相关的关键方面、属性和值。通过结构化地组织社交媒体数据，大规模社交媒体可以在上下文感知的语义OLAP（在线分析处理）框架中进行有效的汇总，并可以在通用的多维社交媒体查询和挖掘框架下进行系统分析，以完成许多任务，例如建模行为模式和发现突发事件以及检测社交欺诈或异常。

Various kinds of social media have impacted billions of users on their ways of obtaining and sharing information across the globe. This creates great opportunities but also poses tremendous challenges on understanding, summarizing, and mining of such data due to its huge volume as well as dynamic and unstructured nature of its text contents. In response to such challenges, this project focuses on text-based social media, proposes a multi-dimensional data structuring approach, which mines unstructured social media data to uncover its hidden multi-dimensional structures. The project investigates principle, methodologies and algorithms for social media structuring, summarizing and mining, and develops effective and scalable technology for multi-dimensional social media data analysis. The principles and methodologies developed in this study can be extended to scalable and multi-dimensional analysis of other kinds of massive unstructured data as well.To conduct effective multi-dimensional social media structuring, this project develops a distant supervision-based methodology with minimal effort of human curation and labeling. It takes data in Wikipedia, Freebase, or other knowledge-bases as references, integrates social media data with the corresponding news or other relevant documents, conducts phrase mining, entity and event discovery and typing, and uncover critical aspects, attributes, and values associated with such entities and events from social media. By organizing social media data in a structured way, massive social media can be summarizing effectively in a context-aware semantic OLAP (online analytical processing) framework and can be analyzed systematically under a general multi-dimensional social media querying and mining framework for many tasks, such as modeling behavioral patterns and uncovering bursty events and detecting social frauds or anomalies.