Characterizing Attitudinal Behaviors in On-Line Open-Sources

On-line public discussions, such as newsgroups, message boards, and other similar forums, are an under-exploited but potentially valuable resource in developing analyses of world events. An effective way of characterizing this large volume of information is to create time-series that represent the subjects, opinions, and attitudes expressed in these sources. Automatically generated “Linguistic Descriptions” then provide natural and easily understood summaries of the behaviors exhibited by these time series. In this extended abstract, we describe preliminary work on the development of a prototype system to implement our concept of Linguistic Descriptions applied to the attitudinal behaviors seen in on-line discussion forums and news sources. Operational Motivation News sources and public opinion polls have long been monitored for indications of public opinion. However, new opportunities exist to understand worldwide reaction to events in the analysis of newsgroups and similar openly available material. Automated analysis of text, until today, has been primarily focused on extracting and understanding the facts represented in text. Yet attitudinal content, though difficult to gauge, is also central to the intelligence production process. We believe that large-scale analysis of multiple sources can help to compensate for the biases and particular viewpoints of any individual forum. Large-scale analysis also provides a base-line against which variations can be discovered and evaluated. The concept of “Linguistic Descriptions” that we introduce in this paper is designed to provide intelligence analysts with a core capability for assessing and interpreting local and global perspectives and their dynamic behaviors. Linguistic Descriptions By a Linguistic Description (LD) we mean constructs such as: “In the last few weeks, Mid-East sources have become strongly critical of US policy in the region.”