Kick off each Monday with the best news and ideas in social media.
Beyond Sentiment Analysis: Canvs Takes Instant Social Temperature Of TV Shows
Judging the success of a TV show used to be a relatively simple matter. Nielsen crunched the viewership numbers and shows survived and died based on the ratings.
The rise of social media and especially its second-screen applications has complicated the picture by adding a rich stream of complementary data. Now television networks and producers can dig into public conversations about their programs to learn much more about how the audience is reacting to the content.
But the blessing of extra data comes with a major chore. TV executives, marketers and advertisers still want simple answers: How did the show do? Did people like last night’s shocking episode of “Fargo”? And that’s not easy to quickly determine from the stream of social media commentary about Billy Bob Thornton and co.
That’s the issue Mashwork founder and CEO Jared Feldman believes he has solved with Canvs, a social TV analytics tool his Manhattan-based startup launched last month.
The solution lies with the natural language processing under the Canvs hood — engineered by Mashwork chief scientist and NYU marketing professor Sam Hui — that allows it to go beyond basic Twitter sentiment analysis of positive, negative or neutral. It’s the product of four years of Mashwork doing much of analysis manually for entertainment clients, Feldman said, building a database of emotionally charged keywords and phrases and their various misspellings.
Such work typically has to be done by hand. Feldman said he knows of one PR firm that sends tweets out to Mechanical Turk for processing. Another network assigns “sentiment interns” to the task.
“The reason that this is such a hard fricken problem is because no one spells things properly,” Feldman said. “People say things like fricken. There’s no academic library or thesaurus on the planet you can use to capture how real people feel about Walter White.”
Canvs In Action
Using its keyword and phrase database, Canvs pulls in all tweets about a show, searching for emotional reactions and sorting them into categories, such as WTF, Terrible, Love, Beautiful, Clueless, etc.
Sifting for similarities it knows, for instance, to group together all the tweets proclaiming the good looks of Lauren London, star of BET’s “The Game.” “Lauren looks good, which is similar to Lauren is so cute,” Feldman said, reading from the Twitter stream from a episode earlier this year. “She looks good, she’s adorable. Really gorgeous, her sexy ass is to die for. You are so fahn, F-A-H-N. Sexy, stunning.
“All the real ways that this very vocal audience that uses a lot of slang a lot of misspellings says that a character is beautiful is listed here. And it’s completely transparent in that if we wanted to we could read every single one.”
The idea is to give a quick overview of performance, and also give users the ability to drill deeper into reactions about characters, actors, topics and moments within the show. From each of those elements, users can pull up the full list of the tweets within.
The timeline-based Moments feature is especially powerful, Feldman said. “It tells you the most important things that happened during your show by looking at massive shifts in emotion,” he said, “meaning if all of a sudden people thinks something was beautiful, Canvs captures that. What was the funniest moment of my show. We captured it, from 7:15 to 7:20.”
BET Is On Board
The BET network, which along with Comedy Central, Spike TV and truTV was a Canvs beta tester, saw an example of that while analyzing an episode of “Being Mary Jane” earlier this year. A brief romantic scene between two of the older characters, played by Richard Roundtree and Margaret Avery, sparked unexpected passion from the show’s mostly younger audience, promoting the network to feature a clip of the scene on BET.com.
“It wasn’t something we were anticipating,” said JP Lespinasse, BET’s senior director of social media. “But after looking at Canvs, we saw that a lot of the emotive tweets that were sent were centered around that scene.”
Lespinasse said he is very impressed with how quickly Canvs allows him to take the social temperature of a show. “Right off the bat, I can get a really good understanding,” he said. ” ‘Did they like it, did they hate it, was there stuff that resonated, Whoa, there’s a grouping that I hadn’t seen before in any other show, let me click on that and see what that means.’ And directly deliver feedback to people based on that.”
Canvs’ audience insights tab — which provides ethnic, income, location, gender and age demographics as well as information about who they follow and what brands they interact with — is also valuable, Lespinasse said. Imagine being able to bring data to McDonald’s that the core audience for “The Game” has a strong affinity for the fast food chain.
Clearly, there’s valuable data to be mined, and Lespinasse said BET is just getting started. Comparing Canvs audience data with Nielsen demographics, he said: “It was interesting to see that there was a bit of a difference between the people talking about the show on Twitter — and specifically the people talking at a high-level, highly invested — and the people who were just passively watching.”
Lespinasse said he’s looking forward to using Canvs during the BET Awards, the network’s summer tentpole event on June 29. Last year, the award show generated 10 million tweets. This year, Lespinasse will be standing by, Canvs at the ready.
“Canvs doesn’t require a data scientist to read it,” he said. “The magic is that I can glance at it and I can get a really, really good understanding about what happened on the show related to the emotions of the audience.”
Mashwork is marketing its product entirely on the TV market and charging on a per series, per month basis, which also gives access to a show’s direct competitors.
Postscript: A previous version of this article incorrectly stated that the 2013 BET Awards show generated 20 million tweets, instead of the actual 10 million.
Some opinions expressed in this article may be those of a guest author and not necessarily Marketing Land. Staff authors are listed here.