Netflix is using real-time sentiment analysis to tell when it's annoyed its users.
There's something about Twitter that encourages complaining. Your flight was delayed? The correct response, obviously, is to inform everyone of the massive inconvenience that has befallen you. Your cable's out? Ditto. Your salmon entrée arrived too cold, your tuna tartare too warm? Ditto.
For users, that tendency can make Twitter occasionally annoying. But for service providers, it can also make Twitter a treasure trove. In general, a healthy percentage of the billion or so messages posted to the service each week contain feedback about companies and their services. And whether that feedback comes in the form of glowing testimonials or incredibly angry screeds, it is instructive. And companies that provide web-based services, in particular, have an obvious interest in tracking that feedback in real time.
One of those companies is Netflix. As the firm, both pre- and post-Qwikster, has expanded its movie-streaming service, it's also had to evolve its customer-service strategy. If Downton Abbey's down, Netflix technicians want to know about it -- and they want to know about it right away, lest their users go all Dowager Countess on them. For Netflix, as for all digital service providers, response times of days or even hours will no longer cut it: Customers expect replies to their complaints that are delivered efficiently, sincerely, and yesterday.
In response to that need, Netflix has been working on a specialized system that identifies outages within its streaming service by analyzing Twitter. The SPOONS system (short for Swift Perceptions of Online Negative Situations) acts as an early warning mechanism for outages, tracking Twitter for those moments when a bunch of people are complaining about the same thing at the same time.
SPOONS works using a combination of term-frequency measurement and sentiment analysis. So: Is Twitter suddenly hosting an increased incidence of references to "netflix"? Hmm, there could be a problem. Do those tweets also feature an increase in negative sentiment, as measured by the emotional polarity of the words they use? Hmm, there's probably a problem. Do those tweets also feature the words "downton" and "fail" and "sucks"? Okay, yes, there is almost definitely a problem.
The SPOONS system hints at the new dynamic developing between companies and their customers -- one in which customer feedback forms and similarly wretched relics of analog consumer life will be complemented, and maybe even superseded, by the implicit feedback of social media. Why take the time to tell a service how horribly it's let you down when that service can see for itself how horribly it's let you down? Companies, you need never wonder whether you're being bitched about again! You can know, with a nice degree of certainty, whether you're being bitched about.
And semantics-and-sentiments are only the beginning when it comes to social media mining. Cal Poly grad student Cailin Cushing, for one, has framed her Master's thesis around the emerging field. Cushing and her team took Netflix's existing SPOONS system and expanded it. They developed eight different outage detection methods. And they rounded those out with 256 different sentiment estimation procedures.
Yep, 256. Let the smack-talking