Grammar-based approaches to opinion mining
ESSLLI 2013 course, Düsseldorf, Germany
On this page are the PDF slides containing my course
notes/presentations for the course I offered at European Summer School on Logic,
Language, and Information (ESSLLI) in 2013. They are only very slightly
edited from what I actually presented at ESSLLI.
Contents
Introduction and motivation
The primary goals of this course were multiple:
- to introduce the surrounding context and underlying concept of
opinion mining to the ESSLLI participants, focusing particularly on
opinion mining/sentiment analysis as the place where social context
meets traditional grammatical concepts.
- to present (in some depth, but limited) a selection of published
advances over the past 10-12 years in opinion mining, focusing on
motivations for including grammatical features.
- to introduce some basic techniques and tools to the participants
through descriptions and demos, including those that enable the use
of rich grammatical constructs.
- and to present my not-unbiased perspective and experience in
opinion mining research and encourage participants to think about
unresolved issues in the opinion mining field.
The course involved a combination of lectures and demos consisting of
five approx. 90-minute sessions. The slides are in five parts which
do not necessarily match the boundaries of these sessions, but roughly
fill up the time with 20 minutes for discussion in the last session,
and two 30-minute demo sessions of Amazon Mechanical Turk and crf++.
The slides
- Part 1: Opinion mining: the larger context
- Sentiment analysis background: product reviews, debates.
- Challenges of perspective, pragmatics.
- Corpus-based social science
- Part 2: Corpora: they are necessary (8mb -- MAY take time to download)
- Corpus requirements for supervised learning.
- Examples of existing corpora (MPQA, JDPA).
- Crowdsourcing for sentiment analysis.
- Part 3: This is not a machine learning course.
- Sequence learning refresher (HMMs, CRFs, etc).
- A little bit of graphical modeling.
- Part 4: You go with the sequences you got.
- Supervised and/or sequence-based techniques.
- Opinion source identification techniques.
- Opinion target challenges.
- Part 5: But...but...GRAMMAR!
- Processing grammatical structure to detect fine-grained opinions.
- Our dystopian future.
Bibliography and background documents
I'm still working on updating the course bibliography, but the original
unedited ESSLLI course application is
here, including a number of references that also appeared during the
course.
Use, attribution, and comments
Feel free to use these materials as long as you give me the
appropriate credit by mentioning me and (even better) giving a link
back to my current home page. Contact me
(spam avoidance: I will leave it to you to figure out how, but my home
page has a very big clue) if you have any comments/requests/corrections
that you want to make.
Last edited 2013 Aug 19, 02:26 CEST