Sentiment Analysis in Practice - an ICDM2011 tutorial
Vancouver, BC, Canada, December 12, 2011
Yongzheng (Tiger) Zhang, eBay Research Labs
Dan Shen, dianping.com
Catherine Baudin, eBay Research Labs
Sentiment Analysis, also known as Opinion Mining, is the computational study of sentiments and opinions expressed in unstructured text documents. With the explosion of user-generated
content on the Web (e.g. blogs, social media, product reviews, and online news articles), it is critical to accurately extract and interpret such sentiments in order to facilitate critical
business processes such as customer support and satisfaction, brand and reputation management, product design and marketing. In this tutorial, we will focus on essential components of a
sentiment analysis system, including sentiment identification, sentiment classification, opinion summarization, and opinion search and retrieval. We will illustrate popular academic
approaches to sentiment analysis and commercial systems and tools. We will also showcase two sentiment analysis systems built in eBay Research Labs to illustrate how to build a practical
sentiment analysis system end to end using product reviews as an example.
The tutorial will be three hours long and is scenario-driven with lots of real examples. In each case, we will discuss the concept, the motivation behind it, and the solutions with real
examples. This tutorial is aimed at researchers, practitioners, and graduate students. We expect the audience to have a computer science background or equivalent. Participants will learn not
only state-of-the-art academic approaches to sentiment analysis but also practical systems and tools.
We are very thankful to Professor Bing Liu for his book chapter Sentiment Analysis and Subjectivity, in Handbook of Natural Language Processing, 2nd Edition (editors: N. Indurkhya and F. J. Damerau), 2010. Many of the contents covered in this tutorial are based on materials in this chapter. We are also
thankful to other authors whose research was introduced in our tutorial. Details can be found in the tutorial slides.
You can download the slides here.
Feedback from attendees
We had a beautiful sunny day in the usually rainy chilly winter in Vancouver :-) And we had a full house of 60 attendees coming from more than 10 countries and majority of them stayed for the
full 3 hours. We really appreciated the great support! We are glad to see the overall opinion on our tutorial is positive, an average score of 8.75 out of 10 :-) The encouragements and
suggestions are so valuable for future improvements. While more deeper sentiment analysis of the feedback in on the way, here are some excerpts from the attendees:
- "He gave an excellent overview of the field, the approaches, the algorithms to use." - from Professor Dirk Van den Poel at Ghent University, Belgium (read full blog)
- "Well organized, consistent across topics, and updated... The tutorial is very good already. Perhaps, shorten the introduction and focus more on the practical examples adding a new one, similar to the lingpipe step-by-step sentiment analysis tutorial." - from P. Di Fabbrizio at AT&T Labs, Research
- "Slides were well organized and clearly presented. Nice real-world examples were used to illustrate problems and support points. Key applied problems and their solutions were covered in this tutorial. Questions from audiences were well addressed during the talk... Besides the general research solutions, experience on any engineering challenges is also encouraged to share, including learned lessons, open source tool to recommend, etc." - from S. Guo at AT&T Interactive
- "I like the breadth of the review of sentiment analysis." - from Professor Hayamizu at Gifu University, Japan
- "It covered all general aspects of sentiment mining, clearly presented the area to newcomers, and provided an overview to researchers in the field reminding us of aspects that still need work." - from R. Hobeica at the American University of Beirut, Lebanon
- "Very organized, very easy to understand, and very helpful... Should spend more time on supervised learning methods such as SVM." - from T. Wang at University of Alberta, Canada
- "The tutorial provides information about practical solutions, not just abstract theory." - from Professor Ad Feelders at Universiteit Utrecht, Netherlands
- "It was clearly structured and easy to understand... I would prefer to get more technical information e.g. algorithms." - from Dr. F. Piazza at Saarland University, Germany
Back to [publications]
First created: 29-Jan-2012. Last updated: 06-Feb-2012.
© 2001-2012 Yongzheng Zhang. All rights reserved.