Q-Phrase: Contextual Relevance Tech Unlocks Deeper Value from SERPs
I first encountered Andy Miller in comments on Google's Orion purchase, where he aggressively marked his territory:

"I'm not sure why everyone is all fired up about Orion, My company, Q-Phrase, has had a superior product in the market since 2004.

ConceptQ Pro does everything Orion claims and then some. If interested, check out www.q-phrase.com."

It took me some time to really dig into the Q-Phrase offering. About 3 months to be exact. I put CQ Web on my PowerBook last weekend and it pretty much blew my mind. Seriously.

I've long been a fan of Ask's Zoom function (Andy Beal was a fan from the get go), as it enables users to dig into concepts.

Not every searcher's looking to drill into concepts obviously, but the zoom function brings an element of *discovery* - of the unexpected, yet relevant result - back into the search process.

CQ Web takes Zoom many steps farther - it unlocks the "story" of an information corpus (whether that's SERPs, LiveJournal or texts that you point the tool at) by parsing out keyphrases and key concepts.

Point it at the whole body of Moby Dick and presto - instant Cliff's Notes.

My initial reaction is that CQ Web - as it currently stands - won't appeal to a mainstream audience though. It requires a download and it's far slower than a straight Google search.

That said, the way that it extracts *concept understanding* out of indices is a wonder to behold, and it could be a strong tool for researchers - both academic and business. Esepcially in that CQ Web enables concept mining into MySpace and LiveJournal.

It could be a great tool also for PR folks seeking to understand an industry's media rapidly, and JP Sherman's currently investigating its competitive intelligence applications.

I, as is my wont, badgered them some about an API. There are MANY companies I've interviewed that could use CQ Web in some form or fashion...

Q-Phrase is launching a new product shortly, and in the interest of better understanding their new offering I asked Andy Miller and Q-Phrase Chief Software Architect Danny Espinoza some questions about their technology.

How is Q-Phrase funded?

Miller: As a privately help company we do not disclose our financial position, however, I can tell you that we are funded by a group of Angel Investors that are committed to making Q-Phrase successful.

What is your definition of contextual relevance?

Espinoza: Contextual relevance is a measure of how a keyword or phrase relates to concepts in which they appear. Take the term “IPTV” for example: in most conversations, IPTV is understood to mean Internet Protocol Television. But if you search for this term in Google, you will find that Iowa Public Television (which is also referred to as IPTV) is a highly weighted link, at least according to PageRank. Now, clearly IPTV is indeed important to their customers in Iowa, but 99% of the people searching for IPTV are probably not looking for a regional television station. The problem here is a lack of contextual relevance.

>Contextual Relevance in ConceptQ
Note that our ConceptQ and CQ web products, do in fact use contextual relevance to determine the importance of pages returned by a given search. For example, CQ web’s analysis of the content on the pages returned by a search for “IPTV” identifies “internet television,” “internet protocol,” “digital video,” as some contextually relevant terms, Thus, the Iowa page is filtered out from our results, because its content does not match the contextually relevant baseline formed by the web’s content relating to IPTV.

>Contextual Relevance in Quece
(ommited til next week - G :)

We think that both of our approaches to contextual relevance, statistical in our ConceptQ and CQ web products and semantic in Quece, are important to solving the problem of understanding the concepts that define many search queries. And, it's only in understanding what search results mean that can we try to improve search relevance.

Q-Phrase API ever available?

Espinoza: We have considered APIs for our ConceptQ product line. ConceptQ looks for contextual relevance in any textual data, and so we allow web searching, site crawling as well as local document review as “sources” for data to analyze.

A natural fit for an API would be a “source” SDK that allowed developers to program their own textual “data gathers” (e.g. for a proprietary database or interfacing to a “deep content” web resource) to pass to ConceptQ for analysis.

What were you envisioning when you included MySpace and LiveJournal as data sources?

Espinoza: Statistical analysis on textual data requires a large enough sample set to determine a contextual relevance baseline.

MySpace and LiveJournal are in essence huge data warehouses of highly topical data, specifically conversations about what today’s youth (in general) are talking about. We created plugins to allow our ConceptQ products to analyze this data as an experiment, but the truth is that most of these conversations are transient in their mentioning of specific terms.

This means that as a statistical sample, more work needs to be done on the filtering side (e.g. using “cultural” stop word lists) to extract real meaningful content from this data. We’re working on it, but including MySpace and LiveJournal was an important first step in getting our heads around the value of this information.

Who’s been digging into these two the most and why?

Espinoza: We haven’t done a formal survey of their use, but we assume most of the interest in analysis of these sites is currently in sheer entertainment value.

Of course, there is a huge potential for real-time marketing research with regards to data mining on these sites, and that’s something were also looking at.

My two queries (brookers - a YouTube “celebrity” with a MySpace page and iPod didn’t really reveal anything...)

Espinoza: That’s not surprising given that these terms are perhaps too vague for extracting meaning from the typical social networking page. Narrowing the focus a bit (e.g. “Paris Hilton” on LiveJournal) can yield some fascinating concepts and conversations.

And as I mentioned earlier, we need to do some work on filtering to improve our ability to get meaningful information from these sites that are generally comprised of pure conversation. Leveraging Quece may be one way to do that.

Shopping applications for Q-Phrase?

Miller: Yes. For example, as we further develop Quece, adding a voice recognition component, we are exploring Quece’s integration with e-commerce/auction sites...

(...oops! that's enough about Quece this week ;)

I have a call later today with Andy Miller and Danny Espinoza, and will have more for you next week regarding their *latest* contribution to understanding.

(toot toot: I played match maker for the Pre Found/Q-Phrase partnership :)
Garrett French - MarketSmart Interactive | permalink | +marktd |