The ongoing civil war in eastern Ukraine provides an example of a situation that social media analytics could be applied to -- for example, by investment portfolio managers looking to evaluate risks to their corporate bond holdings in companies whose operations might be affected by the conflict. But it also shows why social media sentiment analysis can be a difficult task for organizations to successfully pull off.
There's no shortage of news organizations and commentators offering information about the Ukrainian crisis that might be relevant to bond investment teams, both through traditional channels and posts on Twitter and other social media outlets. The challenge, though, is to do systematic sentiment analysis of news stories and social media feeds that assess new developments and point to their possible ramifications for different types of investments. It could be safe to leave money in the bonds of defense companies but not those of natural gas suppliers, depending on the details of what's happening on the ground.
Those are the kinds of questions that Capital Market Exchange grapples with on a daily basis. The Boston-based company was founded three years ago to provide bond investors with insights derived from social media and business news posts. Sarah Biller, its president, said the key to making the process work is not only to look at whether a post is positive or negative, but also to understand more deeply why it was written in the first place.
"In the investment community, [sentiment analysis] is more often not actionable," Biller said. "We went with the idea that if we could ascertain the second layer of sentiment -- not just what's positive and negative, but what drove experts to reflect positively or negatively -- we could model that."
Working with a curated set of analytics data
Instead of looking at the entire stream of Twitter data on a topic like the strife in Ukraine, or every business news publication, Capital Market Exchange gathers information from a curated list of what it considers to be influencers in the bond market. That helps the company put their comments in context, Biller said. It then feeds the data into a homegrown analytics engine, which is partly based on the R programming language. Customers get access to the analytics output through a Web-based dashboard.
Biller said the company had to build its own system for analyzing the data because there weren't any commercially available options that fit its needs. One common problem on any social media analytics project is how to score posts that may include sarcasm, irony or other counterintuitive forms of language. That problem is compounded in the investment world, where commentators often use unique jargon that off-the-shelf sentiment analysis systems don't understand.
For example, Biller said investors typically want to stay away from "rich bonds," ones that are trading at higher prices than financial analysts think they should be. But, she added, most commercial social media analytics engines are likely to interpret the word rich as a positive reference.
"There are tons of off-the-shelf sentiment products, but we found that we couldn't use them because of the specific word choices the investment community uses," Biller said. It's a must, she noted, to "make sure the technologies are not picking up incorrect nuance."
Looks do matter in social media analytics
Another important step for Capital Market Exchange is visualizing the data. That may not seem like a core piece of social media analytics, but the company's clients need to be able to look at the information in the dashboard and quickly discern what it shows. "You have to display the data in a way that makes sense to the investor," Biller said. For example, the dashboard includes "a lot of color and blinking lights" to help make things clear for users.
Social media sentiment analysis isn't an exact science, and Biller said clients use her company's findings only as one piece of information in their decision-making processes -- or, at least, they'd be wise to. "We aren't perfect," she acknowledged. "I wish we had a crystal ball, but we don't."
In the end, how successful Capital Market Exchange is at analyzing the data it collects will be judged on the basis of how its clients do financially by using the findings, according to Biller. "It's a black-and-white world we work in: Do we make investors money or not? Ultimately, we want to give them enough information that they can better anticipate what's going to happen tomorrow."
See how social media analysis software can help tease out useful information
Unclear ROI on social sentiment tools holds back adoption
Why all the buzz around social media analytics?