|
|
DescriptionIn this work we will develop techniques that allow for better predictions of the attitudes and actions of individuals and groups based on the ideologies and sentiments expressed within online virtual communities. This will be done using a mix of social science and computer science methodologies, including public opinion surveys, computer-mediated experiments, event data, and content and social network analysis of material drawn from virtual communities. The connection between online presence and real-world behavior will be developed using choicetheoretic social science modeling that is informed by recent developments in social science models of culture and action.This research responds to the topic #1 of this BAA, "Understanding Human, Social, Cultural and Behavioral Influences" by, among other things, (d) providing methods to transform "ethnographic findings about society's self-perceptions and decision-making skills into frameworks that can inform behavior models" and (f) "measure public opinion through both direct and indirect means in both conflict and non-conflict environments in real time". We will develop methods and models that will take ethnographic descriptions (including selfdescriptions) from online and offline text, and make predictions about the future actions that will be taken by the groups that are represented by such text. We will also develop models of relationships between offline social research data and online content, so that when one form of data is not available, the other can be collected without significant loss of information. Technical ApproachIn this collaborative research, we've planned to utilize both classic social science research methods and social media research studies to investigate how representative activities occurring in the blogosphere are in a corresponding group of the physical environment, how the findings obtained from the blogosphere correlate to those obtained via classic research methods, and how the findings from the blogosphere can help us understand physical group culture and intentions, and anticipate a group's possible actions.In particular, we investigate
In this research we use a multifaceted methodology with two major types of investigation: The first is an empirically-driven attempt to compare conventional social science attitude-measurement techniques, particularly public opinion surveys, to methods based upon analysis of content on the web, particularly information found in weblogs (blogs), and both to contemporaneous data on political events involving the populations reflected both in the surveys and online content. Given that there is only a small population of enthusiastic bloggers (albeit growing fast), it is interesting and important to know how representative activities occurring in the blogosphere are and how the findings obtained from the blogosphere can help understand group culture and intention, and anticipate a group's possible actions. For example, it is well understood and documented that young and educated Muslims are active in blogging or other online discussion forums, but what relationship does this have to with the actions of Muslim extremists on the ground? The second type of investigation comprises efforts to generate theoretical models that can translate information found via web content into accurate predictions about behavior on the ground. To do this is a robust fashion, the latest social science theories of culture and action must be adapted to work on the specific form of content found on websites in general and blogs more specifically. Theories of culture and action are the latest general paradigm shift in social science predictive analysis, and combine elements of traditional social science rational choice theory with the incorporation of cultural into the preference and beliefs structures of agents. As such, they run the gamut from behavioral economics to psychological decision theory, and they are uniquely suited to the types of analysis that are necessary to show how analysis of web content can lead to accurate predictions about future terrestrial events. Related Publications
Project MembersArizona State University, Data Mining and Machine Learning Lab (DMML)
University of Hawai'iAcknowledgmentsThis project is funded by: ONRContact InformationTop Last Update: Sept. 10, 2009 |