A Double Dip of Sentiment Analysis from Thomson Reuters Thomson Reuters may be breaking new ground in the sentiment analysis arena. The company's toolset now comes complete with analytics for both financial services and investment-relation pros.

Giving Trading Machines Market Insight

If you're a Thomson Reuters customer, the company bets you want three things about content feedback:

  • To know what the tonality is
  • To gain insight
  • To alter the message if needed 

Sounds pretty solid, right? Enter two forms of analytics: Using machine reading of Reuters news content in order to help investor trading, the analysis of sentiment feeds into real-time trading execution systems. The accuracy, the company claims, is about 80%.

Behind the curtain, Reuters machines assign numerical “sentiment scores” to words or phrases which are then processed to give an overall positive, neutral or negative score. These scores can then be added together to calculate the overall sentiment for a company, sector, index, etc. 

Next, that information is automatically tailored for PR audiences who want to graphically see how the market sentiment relates to stock price movements.

"The pros we work with don’t have to read every tweet or blog post, they can get a read of the overall sentiment analysis and directionally it‘s correct,” explained Greg Radner, global head of PR Services at Thomson Reuters. “And then if they see some outliers they can quickly dive in and read individual posts or comments.”

The Future of Sentiment Analysis

At this year's Sentiment Analysis Symposium, Radner spoke to the instability of the process. The biggest problem to surface, of course, was the percentage of accuracy. "Beyond 80 percent, the law of diminishing returns sets in as it becomes more costly," he said.

Brad McCormick EVP and Director of Digital, Porter Novelli, agreed. "One of our tools measures to 90 percent, but we need huge amounts of data to get to that threshold, or 150 conversations per day. One of the only times we were able to do that was for mentions of the Iraq war in the New York Times."

The next step for improvement, Radner says, is moving from machine-readable to machine-learnable:

The next thing beyond sentiment analysis is understanding what the nature of those conversations are. It’s only so helpful to be able to say something has a positive or negative tone, but that doesn’t itself give insight into the nature of the conversation, into what people are really saying.  Using Crimson Hexagon machine-learnable algorithms, ThomsonReuters’ Thomson ONE Public Relations workflow platform users can identify the conversations underway, aggregate them, and categorize them into different buckets. Users can train the engine with a little manual processing on the front end about how to categorize posts, and then have it happen automatically going forward.

The company's double dip launched earlier this year after being tested for six months. If you're ready to take the plunge, head on over.