IBM's Watson mines Twitter for sentiments

IBM's new Insights service harvests data from millions of tweets and uses Watson to analyze them for sentiment and behavior

IBM Watson
Clockready (CC BY-SA 3.0)

Last year IBM and Twitter announced a partnership that seemed like an unlikely pairing at first. What did the microblogging service have to offer the legacy IT company that's reinventing itself as a data-and-services outfit?

The answer: IBM planned to use Twitter as a source of information for its Watson analytics services and enable businesses to mine Twitter for sentiment and behavior data. The fruits of that partnership, Insights for Twitter, is now available as a service hosted on IBM's Bluemix PaaS.

Insights works like most Watson-powered services in the Bluemix catalog. Users submit requests via an API -- in this case to search a store of harvested Twitter data updated in real time and going back as far as November 2013.

IBM's Insights for Twitter IBM

IBM's Insights for Twitter uses Watson to perform sentiment analysis on tweets, although the accuracy of the results is still a work in progress. Here, "legal" is highlighted as a positive sentiment term, but the context is more neutral.

The requests can be as simple as a search term, or they can involve sentiment analyses. For example, a company might seek out all the tweets about a given brand that are positive, negative, neutral, or ambivalent (equally positive or negative).

IBM offers a sample application, although the results hint that the accuracy of Watson's sentiment analysis is still a work in progress and will perhaps need to be tuned over time as Watson ingests more Twitter data. Searching for "Microsoft" using positive sentiment analysis turned up many terms that, in context, didn't really have a positive sentiment. For example, "legal" was considered a positive sentiment term, though in the context of a tweet the word was actually more neutral. (It's also not yet clear how well sentiment analysis works in anything other than English.)

Other parsings of the Twitter stream are easier to execute and more accurate -- such as for language, with over 20 languages currently supported.

IBM claims that Insights does more than merely leverage the data in Twitter, it combines that data with other analytics, "such as weather forecasts, sales information and product inventory stats,"  processed by way of other Bluemix services.

Right now the Insight service lets users search up to 1 million tweets for free, which is in line with the early stages of all IBM's Watson-powered services being offered for free while IBM works out a long-term monetization model.

Copyright © 2015 IDG Communications, Inc.