*** Note Special Date ***

******************************************************************************

                                    Seminar


       Department of Systems Engineering and Engineering Management,

                    The Chinese University of Hong Kong

Title:
Mining Chinese Reviews

Speaker:
Kuiyu CHANG
Nanyang Technological University
Singapore

Date : June 19, 2006 (Monday)

Time : 4:30 p.m. - 5:30 p.m.

Venue : Room 513, William M.W. Mong Engineering Building

(Engineering Building Complex Phase 2), CUHK

Abstract:
We present a knowledge-based system to extract product feature-orientation
(sentiment) pairs from online product reviews. Unlike the vast majority of
existing approaches, our system first extracts strong implicit opinions (terms
strongly associated with a product feature, e.g. "heavy" with weight), before
searching for explicit product feature keywords and their associated
sentiments in a review sentence. We call this the "opinion first, feature
second" approach, which incidentally seems to work well with Chinese reviews.
Our system relies heavily on a hierarchical product feature concept model
(ontology) that lists popular feature and opinion vocabulary used to describe
a product. The concept model is built manually using product domain knowledge
and subsequently expanded via HowNet, a Chinese semantic lexicon (Chinese
equivalent of WordNet). Another innovation of our work is the consideration of
the ``neutral" sentiment. We feel that although the "neutral" sentiment may
not be extremely useful in and of itself, it helped improve the accuracy of
the positive and negative sentiment estimates. To the best of our knowledge,
our work is among one of the first studies on Chinese product feature review
extraction at the sentence resolution. Experiments comparing our approach to a
well-known review mining algorithm applied to 500 Chinese mobile phone reviews
shows the feasibility and robustness of our approach. If time permits we will
demo a working web prototype on Chinese mobile phones.

Bio:
Dr. Kuiyu CHANG joined NTU's School of Computer Engineering in 2003.06 as an
Assistant Professor in Information Systems. Prior to that, he served as Senior
Risk Management Analyst for ClearCommerce (Texas), where he helped discover
fraudulent online purchase patterns for a number of Fortune 500 clients,
thereby reducing their potential losses by millions of dollars. In 2001 he co-
founded Mosuma (Texas), a data-mining corporation, which provided data/text-
mining services to About.com (subsidiary of Primedia group) and Akul group,
both based in New York. >From 2000.07 to 2002.04 he worked as a Member of
Technical Staff for Interwoven (Texas), where he led the adaptation of the
company's text classification software product to handle Chinese and Japanese
languages. He was also heading Interwoven's Austin R&D division, where he
supervised a total of 4 graduate interns on developing novel data-mining
algorithms, including clustering and ontology-creation. From 2000.03 to
2000.07, he was the principal architect of the startup Neonyoyo (Texas), where
he co-developed a text-recommendation GUI prototype, which helped secured
Neonyoyo's first customer, and which eventually led to Neonyoyo's acquisition
by Interwoven in 2000.07 for US$70 million.

Kuiyu is a member of IEEE, ACM. He has served as publications chair for
PAKDD2006, program co-chair for Intelligence and Security Informatics 2006
workshop, program committee member for SIAM Data Mining Conference 2006, SIAM
Workshop on Clustering High Dimensional Data and its Applications 2002. He has
co-authored 20 plus papers, of which two received Best Papers Award
(Motorola1996 and ISI2005). He received his Ph.D. in Electrical & Computer
Engineering from the University of Texas at Austin, his M.S. in Electrical
Engineering from the University of Hawaii at Manoa, and his B.S. in Electrical
Engineering from National Taiwan University. He has more than 15 years
experience in machine learning, and has lately been focusing on statistical
pattern recognition techniques and text/web-mining. He is also the inventor of
Probabilistic Principal Surfaces, a novel algorithm to visualize and model
high-dimensional data on a nonlinear manifold such as the sphere. .

　

Note : Cookies and drinks will be available at 4:15 pm.

 _______________________________________________________________________________