Modeling and predicting trustworthiness of online textual information

Modeling trustworthiness of information is a key factor in many decision support systems. Decision makers and citizens are highly influenced these days by information they get from online resources, such as news, online encyclopedias, blogs, forums, and online product reviews. The ease of publishing on the Web, on the other hand, has also allowed nefarious sources to openly express their views and opinions as if they are facts. In this new dynamics, it is important to validate online claims and distinguish fact from fiction. My research tries to address the challenges in modeling trustworthiness of free-text claims.;In this dissertation, I present the need for research on trustworthiness of online information and argue for going beyond structured, extraction-centric approaches to unstructured, textual evidence-driven trust modeling. The overall goal is to provide tools and build systems that would help users judge the veracity of claims. My research focuses on modeling the trustworthiness of sources and free-text claims in presence of unstructured, textual evidence found online. It involves studying multiple forms in which claims can be expressed in free text, understanding the contexts in which they are expressed, and developing ways to incorporate this understanding in building robust approaches to verify claims. I have investigated how to aggregate signals supporting a claim based on the quality of evidence and explored the use of community knowledge (from online forums) to enable Web users to judge the credibility of sources and claims.;Judging credibility of sources and trustworthiness of information is not absolute. It depends not only on various aspects about what is being claimed, but also on the biases of the sources that express such information and the preferred viewpoints of the users consuming that information. To understand these factors further, I studied how human biases affect the credibility judgment of documents and how to build interfaces that help users acquire knowledge in presence of contradictory evidence in favor of or against the claims. In this dissertation, I summarize these findings and insights and propose an automated claim verification system that helps users validate free text claims with trustworthy information from multiple sources.