Welcome to Persephone! This blog is dedicated to promoting the computational sciences, and will feature shorter posts and occasionally longer articles that discuss projects and topics. As way of introduction, my name is Trent Hyer, and I’m a student about to graduate in April. This blog was started to serve as a sandbox for ideas, random writings, and thought-out publications that I want to share with others.
The Calliope Sentiment Parser Series (CSPS) is an educational undertaking in statistical methods for opinion mining. The tools in CSPS are built to help users discover sentiment and opinions in large bodies of text, and has been trained on data from Twitter, Yelp, and Facebook. Biston is the most recent iteration of Calliope, and was released to the public about two months ago, in January 2016.
Why do I mention CSPS here? Well, there’s two reasons. First of all, I’m shamelessly self-promoting my own stuff – I wrote most of the code in Calliope. But secondly, I believe that opinion mining as we understand it is fundamentally short-sighted. Our belief that text is either positive or negative just doesn’t tell us enough to produce useful insights. Calliope SPS has tools that are different than most. Orpheus, the first iteration, gave sentiment ratings from 0 (very negative) to 1 (very positive) and weighted each text according to the likelihood of it being read. Linus attempted to improve on Orpheus by combining tagged data with semi-tagged data (more on that available at Callipe SPS’s site). Biston went even further by using usefulness tags to predict how insightful a given text would be, plus the predicted sentiment of the text.
The best way to get familiar with Calliope SPS is by visiting the site. Otherwise, you’re welcome to download the bistonreport for a more academic introduction.