Vijaykant N had worked on NLP based problem during his stint in the US. He wanted to build a tool for sentiment analysis based on some of his findings. “Instead of doing it myself, I thought it would be a good idea to experiment by bringing in interns to work on this project,” says Vijay. So he started interviewing candidates around December 2013 and shortlisted Jeevith Bopaiah, Sagar M. Rao, Samitha Babu for developing TexSIE.
As there are tons of text that is generated every second owing to the influence of social media, a tool to carry out sentiment analysis becomes valuable. TexSIE was created to solve this interesting NLP problem.
Hiring interns for a real project can be a risk if you fail to plan and have a pragmatic approach. Interns are certainly not as skilled as the industry desires, but they can be moulded. “As founders, we broke up the tasks to granular levels and each one of us took up each vertical to guide. The first task was to make the interns serious about the work that they were supposed to carry out. Interns were given complete freedom, but were always made aware of their responsibility, not just on their academic front, but also from the real world standpoint,” adds Vijay.
The founders of Tationem, the venture behind TexSIE, set up weekly handholding sessions, bringing in the culture of discipline. And of course, after successful completion of each benchmark, they would reward the interns by taking them out for lunch, beer and night out. “The bottom line is — interns should be trusted and guided at the same time. We got to be patient with them to begin with, and see how interns will surprise you by delivering what is assigned to them,” he adds.
TexSIE (link) uses unique algorithm SentiRank, which makes use of hybrid model for evaluation. Elements of statistical methods, learning and linguistic rules are incorporated for better accuracy. A user has options to enter URL or the actual text which he/she would like to analyze for sentiment.
TexSIE has yielded accurate results even on complex sentences. The hybrid model used in TexSIE gives the expected sentiment. Additionally, there is a provision to accept feedback from user and valid feedback is even incorporated for enhancing the accuracy. “Finally comes the crowdsourcing algorithm, which requests users to input a score for a word selected randomly, which after statistical validation is added to the corpus,” adds Vijay.
Within a month of its launch, TexSIE had amassed nearly 14000 hits. “Each user had an option to either agree or disagree with the categorization of the sentiment. As of now, we have an agreement rate of over 95%,” says Vijay.
The targeted audience for TexSIE could be anyone who uses internet for reading news, movie reviews, restaurant, hotel and product reviews. Anywhere there is a large chunk of text and you would like to identify the sentiment without taking the pains of reading it, TexSIE will serve the purpose.
“There is no revenue model directly on this product. We would want to give it out for free. However, an exclusive social media version of the same tool will have a freemium model for revenue,” says Vijay.
The startup is already planning to work on TexSIE 2.0, which will have smarter capabilities such as identifying the area of the entered text (business, science & tech, sports, entertainment, legal et al).
Image credit: Gert Germeraad