In an article published on Sunday — Mining the Web for Feelings, Not Facts– The New York Times gives some well-deserved ink (and clicks) to sentiment analysis, a research area that hasn’t received a lot of attention so far. With the exponential increase in consumer generated media and businesses’ realization that these opinions do have influence on reputation and sales, there’s more interest then ever in finding scalable, accurate and (as much as possible) automated solutions that will be able to take in large amounts of data, and push out insights and trends that can be used as real-time, actionable market intelligence.
The article does a great job at describing some of the challenges of opinion mining and sentiment analysis, but it’s hard to understand the difficulties of creating automated solutions without diving deep in the theoretical underpinnings of the algorithms used by these solutions. If you want to get an idea about what’s out there start by following the sample of links provided below.
What we’re seeing right now are very early steps on a long road, and –although one could think that the solution is at hand when they read in The Times that some algorithms get around 80% accuracy– well… there’s still a way to go. There are many technical and conceptual problems that need to be solved, and they’re far from trivial. Of course, we have a horse in this game, and we’re betting on it. Stay tuned.
- Opinion mining and sentiment analysis, book by Bo Pang and Lillian Lee
- Research papers by Bing Liu, University of Illinois at Chicago (for an overview of the challenges that remained to be solved read his keynote talk at the 5th Annual Text Analytics Summit, Boston, June 1-2, 2009)
- Research papers by Jan Wiebe, University of Pittsburgh
- Research papers by Claire Cardie, Cornell University