Online content is published so fast and furious these days that no one can read it all. Ears are burning at corporate brands, too, because much of that Web content is talking about them. What are people saying? One part of answering that question is to ask how people seem to be feeling about the things they are talking about. Scalable, automated, accurate, sophisticated sentiment analysis is a much sought-after technology that almost no one has really nailed yet.
Jennifer Zeszut was the founder and CEO of Scout Labs, a social media monitoring service acquired by social CRM company Lithium in 2010. ScoutLabs does sentiment analysis, among other things, and Zeszut spoke at the O’Reilly Strata Summit on Big Data this week about the things her company has done that she believes point toward the future of this red hot tech trend.
Zeszut worked previously at eBay and Avenue A I Razorfish. She recently addressed the White House on how to best support tech startups.
Here’s a summary of her advice about sentiment analysis, for companies that build it and customers that buy it. My take-away here? It’s not just technology, it’s also a lot about muscle and brains.
- Harness the wisdom of crowds – scale and let your algorithm learn. Zeszut says that once ScoutLabs got enough scale, across hundreds of companies with hundreds of users at many of them, it was common for one user to manually reclassify the sentiment on an item that was relevant to other users as well. Zeszut says that making sure her algorithm learned from that was key.
- Don’t stop at positive, negative or neutral sentiment. How about noticing people at a moment of indecision? Imagine wishes, caveats, comparisons and preferences. “You should push your vendors for more than just positive, negative and neutral,” Zeszut says.
- Architect for flexibility: if you built a sentiment engine 10 years ago – it won’t work today. If you don’t build for flexibility now – it will not work in 5 years. Several speakers throughout the day discussed the balance between art and science in working with big data.
“Thank you,” Zeszut concluded, “to the Googles and Twitters of the world for giving us metadata, demographic data, geodata. With each new metadata we can filter with easily, we get be more nuanced in what we offer and be more expansive about how we build the algorithms.”
Photo by Pinar Özger for O’Reilly.
