By Rob Knies, Managing Editor, Microsoft Research
Considering the ubiquitous use of search engines to guide users through the Internet, one might be excused for thinking that search functionality has been perfected and there is nothing yet to refine.
Far from it.
Spotlight: Blog post
Eureka: Evaluating and understanding progress in AI
How can we rigorously evaluate and understand state-of-the-art progress in AI? Eureka is an open-source framework for standardizing evaluations of large foundation models, beyond single-score reporting and rankings. Learn more about the extended findings.
Consider a few potential improvements to today’s search engines—and the striking benefits that could result:
- The ability to perform location searches across different languages: What if you could search in English for an address written in Japanese—or vice versa?
- Search results personalized to a user’s intent: If I’m seeking information on the feline species commonly known as the jaguar and you’re trying to buy a shiny new Jag, what if we could both get what we want without having to wade through each other’s extraneous results?
- An entirely new way of determining the importance of Web pages: Computer algorithms are commonly used to determine the best set of results for a particular query. But what if the users were allowed to vote on which results are most appropriate?
These are the subjects of just three of the 16 papers Microsoft Research has had accepted for SIGIR 2008, the 31st annual conference of the Association for Computing Machinery’s Special Interest Group on Information Retrieval.
This year’s conference will be held July 20-24 in Singapore, and of the 85 full papers accepted for SIGIR this year, Microsoft Research participated in the authoring of 18.8 percent. No surprise there; such results merely extend a tradition that goes back years. For example, Microsoft Research contributed 14.9 percent of the conference’s papers last year, 17.5 percent two years ago.
Such numbers reflect the importance Microsoft Research assigns to enhancing the search experience, as does the fact that four of the organization’s six research labs worldwide have had papers accepted for the conference. Of Microsoft Research’s 16 papers to be presented during SIGIR 2008, eight were written in collaboration with associates from academia.
Eight of the 16 papers were submitted by researchers from Microsoft Research Asia, including BrowseRank: Letting Web Users Vote for Page Importance, authored by Yuting Liu of Beijing Jiaotong University, Bin Gao and Tie-Yan Liu of Microsoft Research Asia, Ying Zhang of Nankai University, Zhiming Ma of the Chinese Academy of Sciences, Shuyuan He of Peking University, and Hang Li, also from Microsoft Research Asia.
The paper postulates that enabling users to vote implicitly on page importance provides more reliable results than current algorithms.
Three of the Microsoft Research paper contributions stem from the organization’s Redmond lab, one of them being To Personalize or Not to Personalize: Modeling Queries with Variation in User Intent, co-written by Jaime Teevan, Susan T. Dumais, and Daniel J. Liebling. They will present work that examines the factors that can affect the variability of the intended use for a query with the goal of building models to predict user intent and to provide more relevant search results.
Microsoft Research Cambridge will be represented during SIGIR by four papers, one emanating solely from the U.K. lab and three co-authored with collaborators.
And then there’s the Microsoft Research India contribution, Crosslingual Location Search, written by Tanuja Joshi, Joseph Joy, A Kumaran, Tobias Kellner, and Vibhuti Sengar, along with Udayan Khurana of Microsoft’s India Development Center.
The Crosslingual Location Search work is an outgrowth of Robust Location Search, a project that enables the identification of geographical locations on a map regardless of variations in the structure of the address used. With the latest work, geographical pinpointing can be extended across multiple languages, even those that require transliteration.
Microsoft Research is serving as a platinum sponsor for SIGIR 2008, but the support doesn’t end there. Li will serve as co-chair of the Posters and Demos Committee for the conference, Dumais is co-chair of the Doctoral Consortium, and Ming Zhou of Microsoft Research Asia, is acting as co-chair of Publicity.
With such participation, Microsoft Research is demonstrating its intention to advance the state of the art in information retrieval, with the hope of making your search results of the future more specific and valuable than ever before.
The papers from Microsoft Research that have been accepted for presentation during SIGIR 2008:
A New Rank Correlation Coefficient for Information Retrieval E. Yilmaz, J. Aslam and S. Robertson (Microsoft Research/Northeastern University)
A Unified and Discriminative Model for Query Refinement Jiafeng Guo, Chinese Academy of Sciences; Gu Xu, Microsoft Research Asia; Hang Li, Microsoft Research Asia; and Xueqi Cheng, Chinese Academy of Sciences
BrowseRank: Letting Web Users Vote for Page Importance Yuting Liu, Beijing Jiaotong University; Bin Gao, Microsoft Research Asia; Tie-Yan Liu, Microsoft Research Asia; Ying Zhang, Nankai University; Zhiming Ma, Chinese Academy of Sciences; Shuyuan He, Peking University; and Hang Li, Microsoft Research Asia
Crosslingual Location Search Tanuja Joshi, Microsoft Research India; Joseph Joy, Microsoft Research India; Tobias Kellner, Microsoft Research India; Udayan Khurana, Microsoft India Development Center; A Kumaran, Microsoft Research India; and Vibhuti Sengar, Microsoft Research India
Directly Optimizing Evaluation Measures in Learning to Rank Jun Xu, Microsoft Research Asia; Tie-Yan Liu, Microsoft Research Asia; Min Lu, Nankai University; Hang Li, Microsoft Research Asia; and Wei-Ying Ma, Microsoft Research Asia
Enhancing Text Clustering by Leveraging Wikipedia Semantics Jian Hu, Microsoft Research Asia; Lujun Fang, Fudan University; Yang Cao, Shanghai Jiao Tong University; Hua-Jun Zeng, Microsoft Research Asia; Hua Li, Microsoft Research Asia; Qiang Yang, Hong Kong University of Science and Technology; and Zheng Chen, Microsoft Research Asia
Enhancing Web Search by Promoting Multiple Search Engine Use Ryen W. White, Microsoft Research Redmond; Matthew Richardson, Microsoft Research Redmond; Mikhail Bilenko, Microsoft Research Redmond; and Allison P. Heath, Rice University
Exploring Traversal Strategy for Web Forum Crawling Yida Wang, Chinese Academy of Sciences; Jiang-Ming Yang, Microsoft Research Asia; Wei Lai, Microsoft Research Asia; Rui Cai, Microsoft Research Asia; Lei Zhang, Microsoft Research Asia; and Wei-Ying Ma, Microsoft Research Asia
Finding Question-Answer Pairs from Online Forums Gao Cong, Aalborg University; Long Wang, Tianjin University; Chin-Yew Lin, Microsoft Research Asia; Young-In Song, Korea University; and Yongheng Sun, Tianjin University
Learning Query Intent from Regularized Click Graphs Xiao Li, Microsoft Research Redmond; Ye-Yi Wang, Microsoft Research Redmond; and Alex Acero, Microsoft Research Redmond
Learning to Rank with SoftRank and Gaussian Processes John Guiver, Microsoft Research Cambridge; and Edward Snelson, Microsoft Research Cambridge
Learning to Reduce the Semantic Gap in Web Image Retrieval and Annotation Changhu Wang, University of Science and Technology of China; Lei Zhang, Microsoft Research Asia; and Hong-Jiang Zhang, Microsoft Advanced Technology Center
Query Dependent Ranking Using K-Nearest Neighbor Xiubo Geng, Chinese Academy of Sciences; Tie-Yan Liu, Microsoft Research Asia; Tao Qin, Tsinghua University; Andrew Arnold, Carnegie Mellon University; Hang Li, Microsoft Research Asia; and Heung-Yeung Shum, Microsoft
Relevance Assessment: Are Judges Exchangeable and Does it Matter Peter Bailey, Microsoft; Nick Craswell, Microsoft; Ian Soboroff, National Institute of Standards and Technology; Paul Thomas, Commonwealth Scientific and Industrial Research Organisation; Arjen P. de Vries, Centrum voor Wiskunde en Informatica; and Emine Yilmaz, Microsoft Research Cambridge
Selecting Good Expansion Terms for Pseudo-Relevance Feedback G. Cao, J.Y. Nie, J. Gao and S. Robertson (Microsoft Research/University of Montreal)
To Personalize or Not to Personalize: Modeling Queries with Variation in User Intent Jaime Teevan, Microsoft Research Redmond; Susan T. Dumais, Microsoft Research Redmond; and Daniel J. Liebling, Microsoft Research Redmond
Search and information retrieval