August 28, 2008Computational metaphor identification tool gives unique insight into blog communities
Bill
Tomlinson
Eric
Baumer
Coinciding with the kick-off of the Democratic National Convention, a team of researchers at UC Irvine's California Institute for Telecommunications and Information Technology (Calit2) and Donald Bren School of Information and Computer Sciences has launched metaViz, a Web site that analyzes metaphors in political blogs to help readers better understand the ideologies behind the posts.
As blogs increasingly become a reliable news medium and an important part of Internet culture, making sense of the large body of content has become a challenge.
Graduate student Eric Baumer and undergraduate researcher Jordan Sinclair, along with Assistant Professor of Informatics Bill Tomlinson, envision this computational metaphor identification tool will allow blog readers to understand particular framing or conceptualizations used in different blog communities.
The idea for the tool came when Tomlinson's research group examined multiple political blogs and noted the pervasive metaphors that often underlie the rhetoric in such blogs. "There are close connections between how we think about something and how we talk about it," Tomlinson says. "By looking at patterns in how we talk about different concepts, metaViz offers potential insights into how we think about those concepts."
An example analysis of a community of blogs that link to, and are linked to by, the left-leaning blog Think Progress found that "terrorism" is an "evil attack" or "evil attacker" and that "religion" is also an "attack."
For someone not familiar with the Think Progress community, these metaphors give a quick introduction to the conceptual framings used in the community's discussion.
metaViz works by mining data from the Web site Wikipedia - a large and mostly non-metaphorical source of text - to find common groups of nouns that are characteristic of certain concepts, or "domains," such as military, sports, or parenting. It then analyzes the Wikipedia text for verbs that tend to be associated with those nouns.
Finally, it looks at various political blogs, and determines which groups of nouns tend to be found with the verbs extracted from Wikipedia. The groups of nouns from the blog and from Wikipedia are displayed next to each other, grouped by frequency, as a way of showing how the word usage in the blog tends to map onto the Wikipedia usage.
This project is the first system to find conceptual metaphors in arbitrary bodies of text and publicly display those metaphors in a readily comprehensible manner.
"The goal of the project is to foster critical thinking, to encourage people to look not only at what is being said on blogs by the words themselves, but between and behind the words," says Baumer. "metaViz is designed to enable entirely new ways of reading blogs, allowing readers to see patterns running through an individual blog or across multiple blogs."
