I think that's a brilliant idea. The point I'd like to make is that a combination of data from different channels, would work the best. While for instance, Twitter could be considered as the massive public view on some product, Wikipedia data would be seen as an input about more professional individuals with more accurate information. I think this is the place to point to a recent paper by Osborne et al: Bieber no more: First Story Detection using Twitter and Wikipedia, where they have used such combination to detect "First Stories".


On Thu, Nov 8, 2012 at 10:06 AM, Finn Årup Nielsen <fn@imm.dtu.dk> wrote:



Kerry Raymond: "A really exciting result would be the ability to predict stock price movements from WP editing behaviour!"


I am actually funded by a project where we are trying that. We have looked a bit on Twitter sentiment (like everyone else is doing), but now also do Wikipedia sentiment analysis for companies.

You see an example here for the Lundbeck pharmaceutical company:

http://rb.imm.dtu.dk/base/c/Lundbeck

The plots are for Wikipedia sentiment through time, Twitter sentiment through time and stock price (plots not aligned temporally).

Lundbeck had bad publicity last year. One of their drugs was, without their acceptance, used for executions in United States. There is a drop in Twitter sentiment in regard to that issue -- and also a slight drop in Wikipedia sentiment. It is unclear to me whether the stock price movement is related to that media issue.

I have not completed the analysis. But you see some further companies here http://rb.imm.dtu.dk/base/c/ Mostly it is only the Swedish and Danish companies I have run through the sentiment analysis.


Finn Årup Nielsen

_______________________________________________
Wiki-research-l mailing list
Wiki-research-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wiki-research-l



--
.t