Understanding Your Audience With Sentiment Analysis

The Web is an amazing venue for all types of analysis. No, not the kind on the leather couch…more of the statistical variety. Applications like Google Analytics measure web site traffic while data visualizations depict complex statistical evidence. But is it possible to analyze how people feel while using web sites?

Thanks to Sentiment Analysis, we are now able to scratch this qualitative surface. Also known as opinion mining, sentiment analysis is used to evaluate the emotional state and intended emotional communication of a speaker, writer, or social network user (like those of us plugged into Facebook). The term was first used in the study of linguistics, but is now being used to determine how products, events, and issues are being perceived online.

Think of sentiment analysis as litmus tests for the masses. For example, widgets and other web applications can aggregate and display information from social networks to decipher the emotional temperature of the general public towards an issue, formulating a sort of quantitative zeitgeist if you will. Responses towards specific issues can also be tracked on Twitter, or various vendors can create tools for a specific site that gauge user response.

Utilizing sentiment analysis hinges on the Appraisal Theory, which is the idea that emotions are extracted from our evaluations of events: the same event can be looked at in many ways by different people and elicit different reactions in each of them. This presents a variety of opportunities and challenges when trying to gauge temperaments en masse.

So how does it actually work? One popular method is to look for valence words, which can be used to determine the intended sentiment. Intensifying words, like very and most, indicate the strength of an expressed sentiment, while modal operators like might, could, and should weaken intensity.

And how does all this relate to the Tagged Tanakh (TT)? We’ve actually built sentiment analysis functionality into the prototype through user moderation of content contributed to the TT. By ranking comments up or down, and then qualifying those choices with attributes like useful, funny, or inappropriate, users will provide an indicator of how they feel. For example, if 80% of remarks on the Tagged Tanakh are voted up, that would indicate a generally positive sentiment about a particular remark or tag.

Can emotion ever be accurately converted into numbers (aside from the classic less-than-three heart, <3)? Is Sentiment Analysis an art rather than a science, or useless altogether? We’ll be able to get a better idea of this notion if you share your feelings (pun fully intended) in the comment section!