Our Study of Sentiment Analysis

[ An Industry-Academy collaboration project executed with the software jiant Techmahindra. Students who worked in this project include (i) Saprativa Bhattacharjee, (ii) Anirban Das, (ii) Sanjay Sarkar, (iv) Srijani Pal]

"What others think" is often important to many of us. This is also a major factor influencing the decision of the management of a manufacturer or a service provider in adaptation of the existing policies or framing of new strategies. Now-a-days people express their sentiments over the internet through Twitter, Facebook, Blogs, Forums etc. Analysis of public sentiments over certain specific topic or product is very important to various decision-making bodies. Such sentiments are often categorized as +ve or -ve. However, such a simple categorization is too coarse to its effective use. Categorization of sentiments over a finer scale should help the decision-making authorities in a great way. On the other hand, analyzing these sentiments manually is a difficult task considering the volume of such sentiments spread over the Internet. Thus, the need for some automatic sentiment analyzers is now-a-days growing rapidly. Research and Development works towards such analyzers essentially require some training and test data. In a recent attempt, we developed such training and test data for sentiment analysis in the domain of Cellular Mobile Services. This sentiment data is annotated based on the following taxonomy and the degree of sentiments of these samples is scaled in the set {-2,-1,0,+1,+2}. Also, we developed a baseline system for their automatic analysis. Please click here for web-based demo.

Click to see some sample data

Interested researchers may contact at ujjwal [at] isiscal [dot] ac [dot] in

Back to Ujjwal's main page