Social media intelligence and analytics provider ViralHeat is making its sentiment engine, which it claims as one of the biggest repositories of sentiment data on the market, available for free to developers as an API. The company has been building that engine in conjunction with its agency and big-brand customers (think the likes of Dell) the last few years, and is hoping that the move will open the door to new applications of sentiment analytics, as well as deliver benefits that will profit its paying clients.
“The key for brands and agencies is sentiment,” says CEO Raj Kadam, and ViralHeat got started down that road with a keyword dictionary approach to analyzing social media that proved disappointing. It led to a lot of neutral vs. positive or negative conclusions, and accuracy wasn’t a strong suit. That’s when it turned to its clients to take things up a few levels. “We scrapped that first approach and started building a really large-scale machine learning cluster focused on speed – we get hundreds of millions of mentions a week – and also on accuracy,” he says. Today, the technology runs mentions through its sentiment cluster and gets a sentiment score back, and from there humans play a role in further assessing the text and passing it back to continually train the engine.
Its speed, scalability, and training are what Kadam considers the features that differentiate its Python-built sentiment web service platform from other vendors in the space, and it’s that same Sentiment API that it’s opening up to others. Kadam says the fact that it can quickly do its work, tagging the sentiment score and its accuracy probability on the fly, is one reason why it can open up the API. “If that cluster was really slow it would take us days and probably a large swath wouldn’t get tagged,” he says. He says the latency on competitive systems is “incredible. It’s like batch processing. You send the data in and wait a really long time to get results. Ours is just completely real time.”
Read more