Consumers are starting to realize the first fruits of semantic search, but they’re not thinking of it as such. They’re just thinking of it as better search. “Semantic search is already here but it hasn’t been this ‘Voila!’ moment,” says Scott Prevost, principal development manager, Bing at Microsoft. “Semantic technologies are making an impact across three different dimensions of search: user queries and understanding what people mean when they type a query, having a better sense of what queries mean in a session when people refine the query, and understanding documents and content better.”

Prevost will be giving the opening keynote at this week’s Web 3.0 conference, and he plans to focus on how the incremental inclusion of more and more semantic data in search is affecting everything from search quality, to relevancy of results, to even speed in the sense that people can complete their tasks or find the information they need faster.

When he thinks of semantics, though, he’s not just thinking of search engines taking advantage of data that’s already highly structured or webmasters and publishers beginning to mark up content in formats like RDFa. “There’s a large role for algorithmic discovery,” he says, and all the major search engine competitors are playing that game. “A lot of the work PowerSet [which was purchased by Microsoft back in 2008] did was based on having machines uncover some of the structure for us so humans wouldn’t have to do all the tagging and curating of data. There’s a larger and larger role for that.”

Semantics Needed for Real-Time and Video Search
Analyzing content on a linguistic level to pull semantic meaning out of it and using that to guide results, as PowerSet aimed at, has implications including in the wild and woolly worlds of real-time search, for which Bing has released its Twitter vertical app, and video search. “Semantics is required for real-time search because there’s so few other signals to help judge what these short bursts of communications are,” he says. There’s an “astounding” volume of information out there in Twitter streams, and it will grow. But, for users who want their search engines to help them as they increasingly rely on those streams for their real-time intelligence of hot topics, “you don’t have that linked structure as you do on the web to tell us which pages are important. You have to look at the social graph and how people are connected, the authority and provenance of these communications, all these have to be factored in. One thing that is helpful is being able to understand something about the meaning being conveyed—to at least detect entities that are referenced and disambiguate them.”

Video search, he says, suffers from some of the same problems as real-time search as it relates to having fewer signals to act as guides to search engines about what is really relevant. “Video gets tagged and sometimes there’s a little description,” he says. “And again what’s interesting is that video has a lot of content and most of it is not indexed.” Tags and descriptions get indexed, he notes. But ultimately one of the keys is to index content in video using technologies to automatically transcribe the speech in video. Having some way of semantically analyzing a transcription to understand what it means will help both in finding relevant videos and a;sp in finding relevant sections of videos. Bing doesn’t have anything in the works here that it’s talking about publicly, but video “is on the radar,” Prevost says.

Another area of excitement in search is mobile search, with all its potential for changing the way people interact with search engines using voice input. “But it’s hard to speak in key words. We want to ask a short question. So semantic technologies can help in terms of understanding what those queries are and allowing more flexibility and a more robust query grammar,” he says. The other thing is there is such little real estate for displaying results that you can’t afford to waste any of that space, he says. “So you have to make sure the snippets are the most relevant thing you can show. You want to provide a direct answer whenever possible, puling information from a database or semi-structured page. Semantic technologies are the things that really enable those aspects of search results.”