Comments on: Do we need sentiment analysis for email?

By: Andrew Lampert

Andrew Lampert — Mon, 09 Mar 2009 23:05:53 +0000

Thanks for the pointer to Jane16 – looks really interesting, and excellent to see some open source tools available in the sentiment analysis space. It’s definitely on my list of tools to download and explore.

I agree there are lots of different domains for sentiment analysis, as you mention, and also different scales of conceiving positivity, negativity and neutrality. I wonder how well Jane16 might work on, say, business email (rather than email in general)? Or helpdesk email? I’m just wondering whether you can restrict the domain somewhat to get more consistent results. I’d be really interested to hear about any results if you do try running your tools over email data.

Thanks for dropping by. I look forward to hearing more about your work.

Cheers,
Andrew

By: 123kuko321

123kuko321 — Mon, 09 Mar 2009 22:02:23 +0000

Well, have a look at http://www.jane16.com, it’s the online sentiment analysis engine( downloadable), and you just evoke in me ia idea to use it for email sentiment analysis as well.

The area is a bit more complex cause there are few types of sentiment. (product, financial, emotional, political….plus I am sure many more.)

If only gmail (I use) has a some sort of plugin interface would be great to try it out.

By: Mary

Mary — Tue, 22 Jan 2008 20:39:33 +0000

Turney and Littman also ran their experiments on 10 million words and get about 61% accuracy at the word level using PMI and 65% using LSA. But LSA is fairly computationally expensive.

However, that’s word level classification and you want to classify documents. Classification of entire documents has better results (more features available), so your desktop might provide enough data (although supervised training, as always, would be better).