Jennifer Zaino
SemanticWeb.com Contributor

Semantic advertising startup Peer39 -- which is backed by $12 million in financing from Canaan Partners, JP Morgan and Dawntreader Ventures -- has been recognized by MIT Technology Review as one of the web start-ups to watch this year. The company launched its platform in June, with the aim of leading the next generation of semantic advertising based on both context and behavior.

Founder and CEO Amiad Solomon believes the proprietary algorithms behind its solution will take care of some of the unplanned issues that arise when ads are matched only on the basis of random keywords. For example, imagine someone is reading a blog about someone's trip to Colorado, which might note that the author took some great photos with his camera.

"Contextual technologies will serve up an ad for cameras, but the problem is that's not relevant," says Solomon. That's just a mention within a much broader story. "Then, you don't relate to the camera ad, and so you don't click on it, so the performance is not great. We are not just interested in specific keywords but looking at the page as a whole and what the page is talking about, and matching to the right advertisement with the page."

At the same time, its technology also parses sentiment. If a story, for example, comments negatively on a particular product or type of product or manufacturer of a product, it won't serve up an ad for that company or offering. It relies completely on its ability to "read" pages in real time and understand the user's engagement with the topic to deliver precisely targeted ads, and eschews the use of cookies or other technologies that might infringe on privacy. "We hope this is revolutionary," says Solomon.

At a high level, Solomon says, the algorithms assign different weights to all the text on the page based on how words occur in different sentences and the connections between the sentences.

"We're not just isolating keywords but the page as a whole and understanding the correlation between keywords on the page and between sentences on the page," he says. Developing the machine learning technology to scale and to update itself automatically was no mean feat. "IT took us a long time to move from the first to the second vertical, he says. "Now we ping net for millions of web pages and analyze them often so the results get better and better automatically. We insisted on doing this automatically and now there are more than 1,000 categories covering entire Net."
As the system learns more, those categories can be further divided into smaller and more granularly exact groups. Today, some categories may be very deep, stretching down multiple level s (e.g. from SUV as a broad category to a particular manufacturer to a certain model to a specific year), right down the waterfall.

Solomon says his company is working with top tier ad agencies around the world, but he declined to name customers at this point. The company is engaged in building its own semantic ad network. He sees a lot of opportunities ahead to expand Peer 39's reach in the world of semantic ads, pointing to a list of topics its team has prepared papers on, including visual information extraction and determining an author's native language by mining a text for errors.

Based in Israel, the company has an R&D team 40-people strong, and is expanding that in New York as it opens offices in the states.

The future, Solomon thinks, will go well beyond looking at random keywords and trying to make matches based on them. And maybe it should. Earlier this month, in fact, a law firm representing attorney Hal K. Levitte filed a class-action lawsuit against Google for its alleged sale of low-quality ads, claiming he got only 668 clicks and no conversions for more than 200,000 impressions on parked domain pages.

"I wouldn't talk about Google," says Solomon. "It's a great company and great technology. But in general the more that you see an ad that is relevant to what you are reading, the more you are inclined to work with it. If it's looking at random keywords or even where you were before, it's not necessarily relevant."