Maarten de Rijke

Information retrieval

Month: October 2015

Tutorial on Click Models for Web Search and their Applications accepted at WSDM 2016

Our tutorial on “Click Models for Web Search and their Applications to IR”, with Aleksandr Chuklin and Ilya Markov has been accepted at WSDM 2016, in San Francisco. This will be an updated version of the tutorials we presented at SIGIR 2015, in Santiago, Chile, and at AINL-ISMW 2015 in St. Petersburg, Russia.

WSDM 2016 paper on Multileave Gradient Descent for Fast Online Learning to Rank online

One of our WSDM 2016 papers is online now:

  • Anne Schuth, Harrie Oosterhuis, Shimon Whiteson, and Maarten de Rijke. Multileave gradient descent for fast online learning to rank. In WSDM 2016: The 9th International Conference on Web Search and Data Mining, pages 457-466. ACM, February 2016. Bibtex, PDF
    Author = {Schuth, Anne and Oosterhuis, Harrie and Whiteson, Shimon and de Rijke, Maarten},
    Booktitle = {WSDM 2016: The 9th International Conference on Web Search and Data Mining},
    Date-Added = {2015-10-12 18:46:10 +0000},
    Date-Modified = {2016-05-22 17:59:17 +0000},
    Month = {February},
    Pages = {457--466},
    Publisher = {ACM},
    Title = {Multileave gradient descent for fast online learning to rank},
    Year = {2016}}

Modern search systems are based on dozens or even hundreds of ranking features. The dueling bandit gradient descent (DBGD) algorithm has been shown to effectively learn combinations of these features solely from user interactions. DBGD explores the search space by comparing a possibly improved ranker to the current production ranker. To this end, it uses interleaved comparison methods, which can infer with high sensitivity a preference between two rankings based only on interaction data. A limiting factor is that it can compare only to a single exploratory ranker.

We propose an online learning to rank algorithm called multileave gradient descent (MGD) that extends DBGD to learn from so-called multileaved comparison methods that can compare a set of rankings instead of merely a pair. We show experimentally that MGD allows for better selection of candidates than DBGD without the need for more comparisons involving users. An important implication of our results is that orders of magnitude less user interaction data is required to find good rankers when multileaved comparisons are used within online learning to rank. Hence, fewer users need to be exposed to possibly inferior rankers and our method allows search engines to adapt more quickly to changes in user preferences.

NWO grant: Searching research data

Good news from NWO today. Frank van Harmelen (VUA), Sally Wyatt (KNAW), Andrea Scharnhorst (KNAW), Paul Groth (Elsevier), Anita de Waard (Elsevier) and I obtained an NWO Creative Industries grant to work on search technology for research data sets. The project will fund three PhD students and combine search with knowledge representation and user studies to come up with new ways of searching and recommending research data sets.

Daan Odijk wins CIKM 2015 best student paper award

Congratulations to my PhD student Daan Odijk for winning the CIKM 2015 best student paper award for his paper “Struggling and Success in Web Search’. This full paper is the result of Daan’s internship at Microsoft Research in Redmond in 2014. In this joint paper by Daan Odijk, Ryen White, Ahmed Hassan Awadallah and Susan Dumais, the authors investigate why some web searchers succeed where others struggle.

© 2017 Maarten de Rijke

Theme by Anders NorenUp ↑