On Sat, Jul 18, 2009 at 4:19 PM, Jennie Lees<trin...@gmail.com> wrote:

> I've been working on commercialising sentiment analysis research,
> specifically tuned to microblogs and social media, and my investigations -
> both academic and talking to potential customers - lead me to believe it
> really is worth doing.

Oh, I'm convinced that the "real thing" is worth doing. I'm just not
convinced that

a. Emoticon counting has any value
b. Real NLP-based sentiment analysis can be done easily in Twitter,
given the way the language used in tweets evolves rapidly.

> Sentiment stuff specifically can be done far more
> cheaply compute-wise than full-scale semantic understanding of language.

The literature I've seen ranges from "simple" Bayesian calculations to
the work that Jodange is doing based on some research at Cornell. You
either buy a lot of RAM for your workstation / laptop and harness your
GPU for the linear algebra, or you stand up massively parallel
"clusters" in the cloud to process moderate-sized datasets. Maybe I'm
trying to solve a harder problem than I should be. ;-)

> The key thing though, to any app developer or startup founder, is *not* to
> rely on Twitter. We've been asked this several times by investors now: what
> happens if Twitter fails? Develop stuff that's platform and network agnostic
> and revel in the fact that there's definitely a ton of interest in the space
> right now - despite some players being around for 10 years ;)

There are quite a few interesting approaches / platforms / startups. I
think in the end the issue of Twitter stability / security /
scalability is a non-issue. Twitter is a *success* and the business
intelligence value of tweets will find a way to support the messaging
platform. ;-)

> --J



-- 
M. Edward (Ed) Borasky
http://borasky-research.net

"I've always regarded nature as the clothing of God." ~Alan Hovhaness

Reply via email to