Thursday, September 03, 2009

Sensing Sentiment

Marketing types would love for technology like ContextSense to come of age so they can target their messages to even greater effect. As of now, the results it produces are only fairly interesting - the more culturally distant the theme of the writing is from mainstream America, the more unreliably (and incorrectly) the sentiment is analyzed. The context and the contextually similar links are therefore not meaningful.

One interesting application may be for the software to scan several chapters of a book and pick out the key themes and "sentiment" expressed in it. That can help a prospective buyer (or reader) of the book determine if it would be worth their while working through the whole book. The same is true about a lot of other material one has to read and assimilate. With so much to choose from and so little time to get through it all, help is always welcome.

Grading essays and case studies may be another area where this can be of assistance. If the theme, context and tags are significantly off the mark then clearly it is not a candidate for an A. Better still would be the ability for users to teach the software their personal preferences and worldview. If done really well, the application could play proxy for the real person, assess sentiment like they might have and even come quite close.

1 comment:

Uma said...

I am one of the 'marketing types' :)
I've also been following a lot of tools and technology around 'sentiment analysis' and context etc...but the more I think about it, the more convinced I am that it can't replace a real person. Well, it may take ages at least...