Misspellings, gibberish, sarcasm and slang - welcome to the world of sentiment analytics
Jinfo Blog

By Tim Buckley Owen

Abstract

New on the sentiment analysis scene is Thomson Reuters News Analytics which claims to offer hard, actionable data. Scale is key for success in this field, as offered by LinkedIn, but be warned, the science is still in its infancy with serious limitations.

Item

Sentiment analysis is in the spotlight again as Thomson Reuters rolls out a new analytical product. But as the claims for social media analytics’ capabilities mount up, can it really do everything that the pundits say?

Exploiting information delivered by social media aggregator Moreover, Thomson Reuters News Analytics claims to offer not warm and cuddly impressions but hard, actionable data. Sentiment, relevance and novelty indicators that capture market opinion can be put to use not just for risk management and human decision support processes, but also in algorithmic trading systems, the company says – and it can even be hosted in Elektron, the firm’s own data and trading infrastructure.

Scale is the key factor for making any application of social media analytics work – and scale LinkedIn certainly has, in the business community at least. So when the United States Council of Economic Advisors was putting together this year’s Economic Report of the President, it decided to enlist LinkedIn’s help.

LinkedIn was able to expand on the formal statistical data by showing that, even through the recession, the industries with the largest volume of employment growth were internet, a clutch of health related industries, oil and energy, IT and (by a big margin) renewables. “It’s just a hint of what we can see by scouring the ‘data exhaust’ of 150-plus million LinkedIn members who share information and insights with each other,” commented LinkedIn’s Scott Nicholson in a blog post.

Even the usually sceptical Economist seems enthusiastic. Official statistics can tell you how many workers were jobless last month, how many had college degrees and how many worked in construction – but not how many know the popular data management software Hadoop, as LinkedIn claims it can.

But there is a catch; as Erik Brynjolfsson of the Massachusetts Institute of Technology told the Economist, for consistency and longevity, you can’t beat the government. “Web companies don’t tend to have the same lifespan as the US Census Bureau.”

Another cautious note was struck by the management consultancy Bain in a report last September on Putting Social Media to Work. While several early adopters had captured real economic value from their investments, many others had poured good money after bad in their attempts to engage customers, it found.

Bain reserved its strongest caveats for sentiment analysis (see also previous LiveWire coverage). “Companies considering hiring social media analytics firms should be advised that this science is in its infancy, with serious limitations,” it warned.

Misspellings, gibberish, sarcasm and slang are among the factors that Bain cites, necessitating time-intensive manual manipulation of the data. The industry will improve, it acknowledges – but right now, it says, few sentiment analysis customers would actually recommend the service to their friends.

« Blog