Students and Professor Create Web Site to Analyze Political Language

Informatics graduate student Eric Baumer and Informatics undergraduate student Jordan Sinclair teamed up with Professor of Informatics Bill Tomlinson to develop metaViz, a Web site that explains the language in political weblogs by comparing it with similar phrases from Wikipedia.
The metaViz Web site is the product of Baumer’s dissertation research in which he worked to develop computational methods of analyzing large bodies of text to find language patterns. Rather than identifying the most common metaphors in bodies of text, Baumer sought to promote critical thinking and reflection about these linguistic patterns and what they might mean in regards to how we talk colloquially about political subjects.
“What do these patterns of language imply with respect to how we think about and conceptualize an election?” Baumer said “This type of critical thinking and reflection, looking at not just what is being said by the words themselves, but also between and behind the words, is what we hope to foster with metaViz.”
MetaViz was developed with experienced political blog readers in mind. Thus, it was designed to compare metaphor results with other users and foster critical thought, not draw conclusions itself. The Web site includes a tutorial video detailing the eponymous visualization tool, the heart of the metaViz Web site that allows users to examine popular metaphors from certain blogs (or aggregates of party-aligned blogs), such as Huffington Post, Instapundit, Red County and the Liberal OC, among others and their comparatively-structured phrases from Wikipedia. Using this tool to open multiple mini-windows of blog analysis allows users to track trends of metaphors in said blogs, categorized in respective domains such as “military,” “science” and “religion.”
In practice, the Web site can be confusing: clicking on a metaphor (with a “‘blank’ is like a ‘blank'” structure) brings up a list of keyword metaphors on the left from one subject and a list of correlating phrases culled from Wikipedia on the other. The process is difficult to describe in print and deserves hands-on experimentation to fully appreciate it. Ostensibly, this process is designed to allow users to compare the charged metaphors in the left column with the neutral language in the right, but due to the complicated algorithms of the language-collecting data, this can result in odd comparisons.
“[One problem] was getting the algorithms that identify metaphors to ‘work.’ I put work in scare quotes here because this is a relatively new topic in the field of computational linguistics,” Baumer said. “And as such there aren’t established metrics to determine how accurately an algorithm identifies conceptual metaphors.”
Baumer and his team are in the process of conducting a user study to further refine metaViz. Responses have been positive, especially for the ability to use metaViz to track trends across different blogs, but users are puzzled thereafter.
“Once they’ve found a metaphor, users tend to have a sense of ‘Now what?’ They can see interesting patterns and trends, but don’t know what to do with them,” Baumer said.
In response, Baumer has considered starting a discussion section to better emphasize the “reflect and think” aspect of the Web site. Users could potentially use the discussion section to highlight interesting metaphors, discuss its limitations and suggest alternative metaphors, increasing user interaction and peer communication.
“Incorporating that process of interpretation and meaning-making is key to designing computational technologies that make the user into someone who doesn’t just use technology, but someone who is actively engaged, both with the technology and with the world beyond the technology,” Baumer said.
Baumer is on track to graduate in spring of 2009, but is planning to leave the Web site running to preserve its use for future political blog aficionados. In addition, some users from other educational backgrounds have expressed interest in Baumer’s research.
“Various people with whom I’ve spoken have suggested computational metaphor analysis of texts such as Jane Austen’s novels, the collected works of William Shakespeare, scientific journals and many others could be quite interesting,” Baumer said, “I’m interested in putting the tools necessary to perform those analyses in the hands of those who wish to do so.”

The metaVis website can be found at http://metaviz.ics.uci.edu.