UCI Team Places Third in Microsoft’s Speller Challenge

Microsoft VP Harry Shum presents award certificate to UCI team leader Yasser Ganjisaffar
Ph.D. student Yasser Ganjisaffar accepts the UCI team's award certificate from Microsoft VP Harry Shum

After typing “compter science” into a search engine, most of us don’t think twice about getting results instead for the corrected term “computer science.” The mechanisms behind that correction, however, are anything but simple.

A team of UCI Bren School students recently delved into this problem by entering Microsoft Research and Bing’s “Speller Challenge” and finishing in third place among 100+ competitors worldwide. The aim of the Challenge was to build the best speller — one that proposes the most plausible spelling alternatives for each (potentially misspelled) search query. Spellers were encouraged to use cloud computing to process the terabytes of data needed for a Web-scale speller.

Advised by Associate Professor Chen Li, the UCI team included Ph.D. students Yasser Ganjisaffar (team leader), Sara Javanmardi and Inci Cetindil; M.S. students Manik Sikka and Sandeep Paul Katumalla; visiting student Andrea Zilio from Italy; and volunteer student Narges Khatib-Astaneh, who will join the M.S. program in ICS this fall. They were recognized during an award ceremony held July 20 at the Microsoft Bing Headquarters in Bellevue, Wash.

“Unlike previous data mining competitions in which we had participated, this challenge needed several months of work just to prepare the data sets and the infrastructure needed for processing queries,” says Ganjisaffar, who with Javanmardi works under the supervision of Associate Professor Cristina Lopes. “It was a great experience to solve a Web-scale problem.”

The UCI team’s program, qSpell, addresses the challenges of Web-scale natural language processing. According to the students’ analysis, roughly 10-15 percent of the queries sent to search engines contain spelling errors or typos. These typos must be corrected for users to find their desired results.

“My research group has been working on topics related to Web search, especially on how to deal with errors and inconsistencies in data and queries,” Li explains. “This Microsoft Challenge provided us a great opportunity to test our techniques and help students solve interesting Web-scale problems. The students worked very hard, and I am proud of what they have achieved.”

— PG