From the 1m40s mark to the 3m20s mark a screen shot of Surf Canyon is used to demonstrate dynamic ranking in action. Professor Joachims then offers DCG analysis of a sample search senario and concludes, at the 5m50s, that, “by being dynamic, and adaptive, you can gain a lot of retrieval performance.” Surf Canyon is then mentioned at the 12m00s mark as an Interactive Information Retrieval Model. The Adaptivity Gain, defined previously as the increase in retrieval performance offered by dynamic ranking over traditional static ranking, calculated from empirical studies done on two collections of TREC queries labeled for multiple intents, is then presented at the 14m10s mark and described as “quite substantial” with NDCG going from 55% to 70%.
“When you think about how much effort search engines are spending to get a 1% improvement in NDCG, this is a lot and could potentially change the upper bound of how good you can get with a single ranking.” – Professor Thorsten Joachims
The data once again demonstrates the extent to which dynamically ranking search results in response to real-time user feedback dramatically improves relevance:
Compared to traditional web search, users presented with dynamically ranked results exhibit higher engagement and find information faster, particularly during exploratory tasks. These findings have implications for how search engines might best exploit implicit feedback in real-time in order to help users identify the most relevant results as quickly as possible.
Enjoy the paper and if you attend the CIKM 2013 conference in San Francisco we look forward to seeing you there.
[Update – 2013-8-15] The paper may also be found on Jaime Teevan’s website at Microsoft Research.
We are delighted to announce that on Friday the USPTO awarded Surf Canyon its third patent, U.S. Patent No. 8,442,973, to be issued on May 14th, for “Real Time Implicit User Modeling for Personalized Search.” This is an important piece of intellectual property, as evidenced by the fact it has been referenced over 25 times by other patents issued to companies such as Google, Microsoft, Yahoo!, IBM and Samsung. The Abstract of the invention is as follows:
A method and apparatus for utilizing user behavior to immediately modify sets of search results so that the most relevant documents are moved to the top. In one embodiment of the invention, behavior data, which can come from virtually any activity, is used to infer the user’s intent. The updated inferred implicit user model is then exploited immediately by re-ranking the set of matched documents to best reflect the information need of the user. The system updates the user model and immediately re-ranks documents at every opportunity in order to constantly provide the most optimal results. In another embodiment, the system determines, based on the similarity of results sets, if the current query belongs in the same information session as one or more previous queries. If so, the current query is expanded with additional keywords in order to improve the targeting of the results.
To commemorate this momentous occasion, we have designed and will be giving away a special, universally loved item of tchotchke: t-shirts! You may check out our design and, should you like one for yourself, submit a request on the sign-up sheet. If you leave your address, we’ll do our best to get you one.
This paper, entitled “Personalizing Web Search using Long Term Browsing History,” was submitted to WSDM in 2011, but only recently came to our attention. The authors, Nicolaas Matthijs from the University of Cambridge and Filip Radlinski from Microsoft, write about “a user interest
profile using users’ complete browsing behavior,” but we thank them for taking a moment to recognize Surf Canyon’s contribution to the field:
Recently, personalized search has also been made available in some mainstream web search engines including Google and Yahoo!. These appear to use a combination of explicitly and implicitly collected information about the user. Many more companies are engaging in personalization both for search (e.g. surfcanyon.com) and for advertising based on user behavior.
Mark Cramer, CEO of Surf Canyon, has been honored with an invitation to present Dynamic Ranked Retrieval at the next Bay Area Search Meetup. Anyone near the offices of eBay in San Jose on Wednesday, January 23rd, at 6:30pm, and who would like to attend, should click over to the Meetup website. Here is an Abstract of the talk:
Dynamic Ranked Retrieval – “Unsticking” the SERP
Search is stuck. And, frequently, so too are searchers. The problem stems from the fact that, as Jaime Teevan described, “Web queries are very short, and it is unlikely that a two- or three-word query can unambiguously describe a user’s information goal.” Search engines respond to this by attempting to divine the user’s intent by exploiting massive quantities of data prior to delivering a static SERP, but the onus of either unambiguously describing the intent, or digging through pages of “best guesses,” often falls upon the user. This can be a daunting task depending on the searcher’s skill and knowledge of the subject matter as well as the clarity of the information need.
Dynamic Ranked Retrieval addresses this problem by creating a fluid SERP that re-ranks results “on the fly” in response to real-time implicit feedback from the user. The strongest signals are the immediate ones, and the half-life of their informational value is extremely short, so waiting until a subsequent query to exploit them is deficient. Adapting the SERP dynamically to a real-time model of inferred user intent built from implicit feedback will maximize relevance and searcher satisfaction without requiring additional effort. We will demonstrate the technology and present real-world data on the impact to IR.
[Update 1/28/2013] The video from the presentation is now available online:
“To be honest, I don’t google anymore. Search engines like Surf Canyon work just as well…”
We would point out that, for many queries, the innovative Dynamic Ranked Retrieval technology that we developed and embedded into our search page significantly improves the search experience over Google by automatically digging out relevant results for the user, but we are flattered nonetheless.
ChengXiang received his Ph.D. in Computer Science from Nanjing University in 1990 and another Ph.D. in Language and Information Technologies from Carnegie Mellon in 2002. His research interests include information retrieval, text mining, natural language processing, machine learning and bioinformatics, and he has published over 100 papers in major conferences and journals in these areas, including five award papers.
For ambiguous queries, conventional retrieval systems are bound by two conflicting goals. On the one hand, they should diversify and strive to present results for as many query intents as possible. On the other hand, they should provide depth for each intent by displaying more than a single result. Since both diversity and depth cannot be achieved simultaneously in the conventional static retrieval model, we propose a new dynamic ranking approach. In particular, our proposed two-level dynamic ranking model allows users to adapt the ranking through interaction, thus overcoming the constraints of presenting a one-size-fits-all static ranking.
Surf Canyon is again referenced, along with our 2009 SIGIR research paper:
We argue that a key to solving the conflict between depth and diversity lies in the move to dynamic retrieval models  that can take advantage of user interactions. Instead of presenting a single one-size-fits-all ranking, dynamic retrieval models allow users to adapt the ranking dynamically through interaction, as is done by surfcanyon.com .
In February 2008, Surf Canyon launched its Dynamic Ranked Retrieval application to rave reviews. As the body of research relating to Dynamic Ranked Retrieval grows, we continue to be encouraged by the potential of this technology to vastly enhance the quality of information retrieval.
The mission of Rank Dynamics is to transform search into a dynamic experience where fluid result pages respond to user actions in real time. We develop Dynamic Search, a real-time contextual search technology. By transforming static lists of links into dynamic search pages that automatically and immediately re-order results in response to user behavior signals, searchers are able to more quickly and easily find pertinent information that might otherwise have remained buried as deep as page 100.