What is Opinion
- Opinions changes over time
Applications
- word mining
- extraction
- summarization
- tracking
- QA
- Cross-Lingual
Fact vs Opinion
- Differece
- circular vs Happy
- "he is a engineer" vs "He thiks that a"
Previous Work
- English: [Weibe et al. Kim and Hovy, Takamura et al.]
- Japanese:
- Chinese:
- Korean:
Corpus Preparation
- Quantity
- Source: News, Reviews, blogs,...
- Granularity
- different source/languages...
Previous Work (COrpus)
- NRRC Summer workshop on Multi-perspective QA
- World New Connection (WHC)
- NTCIR Corpus (Chinese
ZH021
ZH024
ZH026
ZH-27
ZH028
ZH029
- Review from Web (Japanese)
Corpus Annotation
Previous Work (Annotation1)
- NRRC - 114 doc. annotataed
Previous Work (Annotation2)
- Tags
Previous Work (Annotation3) - NTCIR Chinese
^ 205 doc. annotated
- 3 annotators
- XML-style tag
Corpus Evaluation
- How to choose materials? (Agreements...)
- How to evaluate corpus for a subjective taslk
Previous Work (Corpus Evaluation)
- bigger annotators, lower agreements...
Needed for this work
- what kind of documents
- training
- which Language?
Comments/Questions:
- How to construct output for
- extract/arrange opinions from
- only extracted opinion background for some other information
Opinion Workshop - ChenYewLin:
MT?
Opinion Workshop - Prof.Kando:
History
- IR
- EAAT-AAAI2004, Style Workshop (SIGIR2005)
Information (type/...)
Aspect from different communities
- Smae event from different language (News,Web etc.)
Technique
--function words
- puctuations, symbols
- aspects, modalities, tense (would, could, have been,...)
- semantic categpries of verbs
- very low frequent terms
Resources
- dictionaries
- corpus (annotrated)
- taggers, matchers
- test collections/test sets
Target applications?
Possible document collections
- Yahoo!J/T: 知恵袋、知識家
- (still) Newspaper: opinions (authories'/writers')
Piliot Task
- single task
- Or everybody work on own research on same doc.
- very relaxed way starting of can be possibly for topics
- it would be some kind community
Feedback (comments/questions)
- Human assesment:
- corpus:
- contact with give us you can use it in labolatories
- i try ato my best possibilities to use Taiwan
- Gardens:
- we can
- for relevant doc. assessment,
- pool
- participants contribute theri