Friday 21 September 2007

Unsentimental targeting of Sentiment Analysis

Fourteen months ago, I wrote an optimistic column about Corpora and its sentiment analysis software. It can take a body of news stories about any particular subject and tell you whether the underlying sentiment is positive or negative. I thought this would be a boon for marketing and PR types for tracking the market's reaction to their activities.


At the time, the company hinted at an imminent big win. This turned out to be a partnership with Reuters. Being a media type, I had jumped to media-related conclusions. But the company (now called Infonic) has tapped a much richer, but more mammonistic, seam. The two companies have slipped a Sentiment Analysis layer into the data management layer of the Reuters Market Data System (RMDS) used by banks and hedge funds.


Reuters is the biggest news agency in the world. If we had time to read all its output, then our view of the world would be more or less well-rounded. Investors aggregate as much of this information as they can in order to make their buy/sell decisions. But, in the end, all they're care about is whether there's a gap between reality and the market's perception. Trading on these gaps is how they make their money.


Governments, too, are very interested in these perceptions. So that's two markets with more or less bottomless pockets. (Update: I forgot to re-mention pharmaceuticals. The average drug has 70,000 trial reports which can be analysed unemotionally for trends.)


Infonic says that it will include Sentiment Analysis into its own search engine but, judging from the company's commercial priorities, I am less optimistic than I was about us getting our mitts on it.

3 comments:

  1. Any idea about the Sentiment system price?

    ReplyDelete
  2. From memory, I think we're talking 'from' hundreds of thousands of pounds. You'd need to contact the company for details.

    ReplyDelete
  3. The price depends on the level of service, the taxonomies involved , the usage of semantic web, the granularity level, and the accuracy of the automatic classification. it may range from few thousands using known algorithms to hundreds of thousands.
    Another important factor of any sentiment system is its capability to take measurable actions in an automatic fashion on the extracted knowledge to achieve the required outcomes.

    ReplyDelete