SVM Indexing and Processing for Data Retrieval
In this talk, I will first present a recent work on SVM indexing and
processing which was published at SIMGOD 2011, and I will introduce
several ongoing research projects of our data mining group at POSTECH,
such as Novel recommendation for digital TV, Protecting location
privacy of mobile devices, and Online advertizing for sponsored
search. These projects are supported by Samsung Electronics and
Microsoft Research Asia.
In applications such as relevance feedback search system, a query is a
ranking function F learned by a machine learning methodology such as
SVM, and the query result is a set of items ranked the highest
according to F. Processing the query F to find top-k items requires
evaluating the entire data by F. We developed an indexing method for
query of an SVM ranking function, which enables quickly finding top-k
items without evaluating the entire data. Our indexing method,
iKernel, produces overall 1~5% of evaluation ratio on large data sets.
iKernel is currently the only indexing solution which finds exact
top-k items for SVM functions. This work passed the repeatability test
of SIGMOD 2011.
Hwanjo Yu received his PhD in Computer Science at the University of
Illinois at Urbana-Champaign at June 2004 under the supervision of
Prof. Jiawei Han. From July 2004 to January 2008, he had been an
assistant professor at the University of Iowa. After that, he joined
POSTECH (Pohang University of Science and Technology) as a faculty
member. His research areas include data mining, database, information
retrieval, machine learning, and medical informatics.