The New Math of Social Media Research

Automated tracking of social media remains a challenge

By George Winslow -- Broadcasting & Cable, 11/1/2010 12:01:00 AM

RELATED: Media Companies' Social Problem

Faced with the challenge of tracking and analyzing billions of posts, blogs, tweets and other social media artifacts, companies are increasingly turning to a handful of major providers of social media tracking services that use sophisticated software systems based on cutting edge math to crunch through massive amounts of data each day.

Some of computer tracking system are already producing important practical insights into social media conversations that major media companies and the Hollywood studios are using to promote new films and programs. "In the movie industry our clients are already amazingly astute in using our services," notes Jennifer Zeszut, chief social strategist at Lithium Social Media Monitoring, which works with a number of studios. Based on the social media conversations "they can tell four weeks out if something is going to be a big hit and if it doesn't look good four weeks out, they are putting an amazing amount of effort into changing the course of events. It is pretty amazing how predictive some of these tools have become."

As time goes by, these insights could become even more valuable, potentially becoming the basis for improved strategies for promoting new shows or network programming.

Melyssa Plunkett-Gomez, VP of sales and business development at Crimson Hexagon, says that they worked with a client to analyze a new one hour U.S. network show. The show received huge critical acclaim when the pilot was screened in May, but opened with poor ratings and was quickly cancelled in the fall.

In tracking social media chat about the show, they found that it had attracted very positive comments but not a lot of conversation, she notes. "One of the things that this customer learned was that if they had real time data, they could have made investments to help build the conversation and awareness of what was a really good show," she says.

Winning Combos
Over time, providers are also hoping to more closely integrate this social media data with TV ratings and other more traditional metrics. "It is very important to look at everything in an integrated fashion-you can't just look at things in different silos-and we're spending a lot of time blending data from other sources, like Nielsen's ratings" with the work they do tracking social media, notes Pete Blackshaw executive VP of digital strategic services at NM Incite, a joint venture between Nielsen and McKinsey & Company.

Such efforts could provide researchers with a much wider look at a show's performance and provide networks with a way to sell both a show's ratings and its audience engagement. Zeszut notes, for example, that ratings and the amount of social media activity aren't always related.

"One client of ours, Disney, was shocked to see how they're different than the official Nielsen rating" was from the amount of activity the shows were getting in social media space, she says. "They were able to use that to win advertising deals [with companies] that were looking to connect with a very young social media crowd."

At the moment, however, social media researchers are still struggling to solve a number of more basic problems.

Sifting Through A Flood of Sentiment
One major issue is simply narrowing down the flood of information. Scott Centurino, CEO of Crimson Hexagon, which works with such companies as CNN, Dow Jones and a number of major broadcasters and agencies, notes that anyone trying to track the reaction to the movie Catfish with a simple key word search would find themselves crunching through a huge number irrelevant responses about cooking catfish or going fishing.

Figuring out what the relevant results might mean is also difficult. On the simplest level, many companies use social media tracking services to see if their brands are being viewed positively or negatively by consumers-something that is called sentiment analysis.

But this kind of analysis is complicated by the sarcasm, irony, slang, abbreviations, emoticons, poor grammar and bad spelling that widely appear in social media texts. "Computers have to take that very freeform language and try to decide if it reflects a positive, negative or neutral assessment of a brand," notes Zeszut.

To make that determination, social media tracking services draw on a number of different types of research, including natural language processing, advanced statistical techniques and semantic analysis.

Natural language processing, which grew out of work on such areas as artificial intelligence and the automated computer translation of foreign languages, uses complex algorithms to process and interpret texts so the information can be automatically summarized and analyzed by computers.

Sematic analysis uses advanced math and algorithms to try to uncover the meaning by tracking patterns of words. By breaking up the underlying semantic structure of a sentence, the computer determines which of the words expressing emotion are connected to the key word being tracked.

Companies like Lithium Social Media Monitoring, formerly Scout Labs, use semantic analysis while Crimson Hexagon uses what it calls a statistical approach.

Based on research developed by Gary King, a co-founder of Crimson Hexagon and the director of Harvard University's Institute for Quantitative Social Science, their software also uses proprietary algorithms to track the statistical relationships and patterns between words but it relies on users to determine if those relationships express a positive or negative sentiment.

With their system, users are given a number of sample posts and then the software uses their classification to analyze the rest of the material, notes Centurino, who says this helps avoid them imposing their biases on the analysis.

Dr. Nick Koudas, president and co-founder of Sysomos, notes that their platform uses three approaches-natural language processing, statistical and semantic tools and stresses that their product is able to handling a large amount of international data. "You can crawl and collect data from 186 languages in 1989 countries," says Koudas, a well-known researcher who holds more than 20 patents.

Measuring ROI
Koudas adds that they will soon be introducing a new product, Sysomos Audience that will help people to analyze the return on investment they've achieved from social media. "It is important because it will give our clients the means to say social media is important because this is the ROI we've achieved by using this technology," Koudas notes.

As researchers work to improve their software, some users stress the importance of being able to see how the system is analyzing results so they can correct any errors that might lead to misinterpretations. 

CNN, for example, used Crimson Hexagon for its coverage of the State of the Union speech earlier this year and is using the social media tracking service for its mid-term election coverage. While they've been very happy with the results, Alex Wellen, senior executive producer, integrated programming at CNN, stresses that "given the state of the technology you don't want to take the data and just stick it into a machine and have it pop out all the answers. As a news organization we owe it to our audience to play a role in that analysis."

Companies also need to work especially hard to make sure they follow up on whatever results they get. "It is not enough to simply give people social media data, you have to understand how to make it actionable," notes Blackshaw. "If you are going to spend dollars on listening to social media, you have to find a way for that information to bubble up through your organization" so that it becomes more "strategically important" to the company's efforts.