The Huffington Post
Adaptive Semantics

Huffington Post Buys Adaptive Semantics To Keep Up With 100,000 Comments A Day

Next Story

Video: Hands on with the Sony Ericsson Xperia X8

The Huffington Post has acquired its first company in a small cash deal, and it is not another blog or media site, but a pure technology startup called Adaptive Semantics. The two-person startup provides a semantic analysis engine (aka JuLiA) already used by the Huffington Post to help moderate the 100,000 comments published on the blog every day.

Prior to the acquisition, the Huffington Post was already Adaptive Semantic’s largest and only outside investor, buying a 20 percent stake in April, 2009. Adaptive Semantic’s two co-founders, Elena Haliczer and Jeff Revesz, will join Huffington Post to oversee its social news and community technology R&D. The acquisition price was not disclosed

“Technology is very critical to us,” says CEO Eric Hippeau. “In this case, the technology has implications for our content. It makes moderation hyper-efficient.” With close to 3 million comments a month, the only way to moderate them is through automation tools (as well as a corp of about 30 professional human moderators).

Other companies that license Adaptive Semantic’s technology for online comment moderation include CNN, Newsweek, and Disqus. They might have to start looking for other solutions. “We will honor the contracts, but very likely will not renew them,” says Hippeau, who doesn’t want to be in the business of licensing technology to other news sites and services.

JuLiA uses “supervised machine learning,” according to Revesz, to flag inappropriate comments, spam, and abusive language. Humans manually tag a few hundred comments, which then get fed into the semantic analysis engine and applied across every comment. This is an ongoing process so that the system continually gets better and better. Not only can it detect abusive language or hate speech, but it can also help find commenters who may be topic experts.

Beyond comment moderation and making sure readers behave themselves on the site, the underlying semantic analysis technology can help bubble up the best contributions from readers. “I am very confident that we are going to find all kinds of ways to apply it,” says Hippeau.

The Huffington Post very much sees itself as a social news network, and its success is tied to engaging its readers in a variety of ways, from leaving comments to sharing posts across the Web via Twitter and Facebook. It recently started awarding readers badges. JuLiA could help to feature the best comments or to award specific badges. For instance, if a reader leaves a lot of comments on posts about Afghanistan, Iraq, and Hillary Clinton, they could get a Foreign Policy badge. That is just a hypothetical example, but the technology opens the door to those kinds of features.

It also could be applied to article recommendations. “The Huffington Post talks a lot about their social graph,” says Haliczer, “how people are connecting to each other and connecting to content. We can look at the content graph and recommend content to people.” Whatever the Huffington Post will end up doing with the technology, it is important enough that the company wants to own it in-house.

blog comments powered by Disqus