Multi_obj

Subjectivity in languages other than English Objective Summary Tyler Thornblade

A System for Summarizing and Visualizing Arguments in Subjective Documents: Toward Supporting Decision Making Fujii, A. & Ishikawa, T., ACL Proceedings of the Workshop on Sentiment and Subjectivity in Text 2008 Read by Michael Lipschultz

A System for Summarizing and Visualizing Arguments in Subjective Documents: Toward Supporting Decision Making Fujii, A. & Ishikawa, T., ACL Proceedings of the Workshop on Sentiment and Subjectivity in Text 2008 Read by Michael Lipschultz



 

  

  

A System for Summarizing and Visualizing Arguments in Subjective Documents: Toward Supporting Decision Making Fujii, A. & Ishikawa, T., ACL Proceedings of the Workshop on Sentiment and Subjectivity in Text 2008 Read by Michael Lipschultz





ISCAS in Opinion Analysis Pilot Task: Experiments with sentimental dictionary based classifier and CRF model Huang, R., Sun, L. & Pan, L., Sixth NTCIR Workshop 2007 Read by Cem Akkaya

Conditional Random Field (CRF) model –

related to HMM Started with lexicon provided by NTCIR Frequency based: F(character in positive

words) - F(character in negative words) Words aggregate character scores Sentences aggregate word scores

Negation heuristic: flip sentiment following

negation term Weight on opinion holders (how?)

ISCAS in Opinion Analysis Pilot Task: Experiments with sentimental dictionary based classifier and CRF model Huang, R., Sun, L. & Pan, L., Sixth NTCIR Workshop 2007 Read by Cem Akkaya

 Fairly typical features; more notably: Named Entity recognizer Opinion sources are triggers for nearby opinions

 Evaluation Subjective/Objective and Polarity

Opinion holder extraction: F 0.489

ISCAS in Opinion Analysis Pilot Task: Experiments with sentimental dictionary based classifier and CRF model Huang, R., Sun, L. & Pan, L., Sixth NTCIR Workshop 2007 Read by Cem Akkaya

Related works: John: "Opinion Extraction Summarization And

Tracking In News And Blog Corpora Authors: Ku, L., Liang, Y. & Chen, H.Published in: AAAI 2006  Also used concept of sentiment attached to

ideograms  Very different source data for lexicon creation

"Identifying sources of opinions with

conditional random, fields and extraction patterns. Choi et al. EMNLP 2005".  Slightly more sophisticated techniques

Extracting Semantic Orientations of Phrases from Dictionary Takamura, H., Inui, T. & Okumura, M., ACL Human Language Technologies 2007 Conference Read by Mahesh

Idea: nouns acquire semantic orientation in

the presence of adjectives Potts model For each adjective, connect to it nouns that

co-occur For each noun, connect it to other nouns that appear in its gloss  The edge weight is negative if the appearance in the

gloss is followed by a negation term

Extracting Semantic Orientations of Phrases from Dictionary Takamura, H., Inui, T. & Okumura, M., ACL Human Language Technologies 2007 Conference Read by Mahesh

Notable points Unusual model “Unambiguous” adjectives Can accommodate certain (labeled) and

uncertain (inferred) data naturally in the graph representation Can dynamically add new words in cases where noun is novel (but appears in lexicon glosses)

Extracting Semantic Orientations of Phrases from Dictionary Takamura, H., Inui, T. & Okumura, M., ACL Human Language Technologies 2007 Conference Read by Mahesh

 Related works  Tyler Thornblade:Seeing stars when there aren't many

stars: Graph-based semi-supervised learning for sentiment categorization,Goldberg, A. B. & Zhu, X., HLTNAACL 2006 Workshop  Another graph based technique; this is different in that it is

more of a global technique than one used to generate just the lexicon but the model is applicable

 Tyler Thornblade:Taking Sides: Graph-based user

classification for informal online political discourse,Malouf, R. & Mullen, T., Article, 2008

 This is another graph-based technique but one that is very

different and would be hard to apply to this work

 Latent variable models for semantic orientations of

phrases, EACL 2006  Prior work

Deeper Sentiment Analysis Using Machine Translation Technology Hiroshi Kanayama and Tetsuya Nasukawa and Hideo Watanabe, COLING-2004 Read by Shilpa Arora

Adapts MT techniques to sentiment analysis Converts Japanese text to form: [sentiment] predicate ([attributes]+) Examples  (a)[favorable] excellent (lens)  (b)[unfavorable] high (price)  (c)[favorable] problematic+neg (recharger)

Deeper Sentiment Analysis Using Machine Translation Technology Hiroshi Kanayama and Tetsuya Nasukawa and Hideo Watanabe, COLING-2004 Read by Shilpa Arora

 Main patterns  Principle pattern: Identifies sentiment units like "[unf] bad <noun>", "[fav] like <noun>" etc  Auxiliary pattern: Identifies sentiment units in sentences like "I don't think X is good" that produces a sentiment unit with negative feature  Nominal pattern: This pattern is used to avoid a formal noun (nominalizer) being an argument  Evaluation

Deeper Sentiment Analysis Using Machine Translation Technology Hiroshi Kanayama and Tetsuya Nasukawa and Hideo Watanabe, COLING-2004 Read by Shilpa Arora

Key points Deep parsing  Is not confused by sentences like “I hope that X is good”  Recognition of double-ga affective adjectivals  Wa and ga for sentiment disambiguation? Nice annotator (SentimentAnalyser)

Exploring in the Weblog Space by Detecting Informative and Affective Articles Ni, X., Xue, G-R., Ling, X., Yu, Y., & Yang, Q., WWW Conference 2007 Read by Danielle Mowery

Document-level analysis of informative vs.

affective articles in the domain of Chinese blogs Three algorithms Naïve Bayes SVM Rocchio

Large number (20,000) of feature

candidates evaluated with Chi Square

Automatic Detection Of Quotations in Multilingual News Pouliquen, B., Steinberger, R. & Best, C.,Proceedings of RANLP 2007 Read by Matt McGettigan

Four types of target sentences

1. Tony Blair said "We stand ready to support

you in every way." 2. "We stand ready to support you in every way," Blair said. 3. Tony Blair visited Iraq… He said "We stand ready to support you in every way." 4. Tony Blair visited Iraq… "We stand ready to support you in every way," the British Prime Minister said. Pattern-based matching across 12

languages, with both language specific and independent patterns

Automatic Detection Of Quotations in Multilingual News Pouliquen, B., Steinberger, R. & Best, C.,Proceedings of RANLP 2007 Read by Matt McGettigan

Evaluation

Precision  81.7% correct, 17.5% incomplete, and 0.8% wrongly assigned Recall  Overall, 13%  76% of all quotes were explicitly unrecognizable  Of the remainder, 54% were recognized Both measurements done over very small test

sets

Automatic Detection Of Quotations in Multilingual News Pouliquen, B., Steinberger, R. & Best, C.,Proceedings of RANLP 2007 Read by Matt McGettigan

Related works

MemeTracker  Simpler system

Searching for Opinions by using Declaratively Subjective Clues

N. Hiroshima, S. Yamada, O. Feruse & R. Kataoka; ACL 2006 workshop on Sentiment and Subjectivity in Text.

Extract a large number of webpages, extract

subjective sentences and finally “declaritively subjective clues” (DSCs) via annotation study Number of DSCs > threshold used to indicate subjectivity Grouping of DSCs into “semantic categories” E.g. “appearance” Helped recall and F-measure greatly at some

cost to precision

Other interesting ideas

Searching for Opinions by using Declaratively Subjective Clues

N. Hiroshima, S. Yamada, O. Feruse & R. Kataoka; ACL 2006 workshop on Sentiment and Subjectivity in Text.

Results:

Searching for Opinions by using Declaratively Subjective Clues

N. Hiroshima, S. Yamada, O. Feruse & R. Kataoka; ACL 2006 workshop on Sentiment and Subjectivity in Text.

Related works Tyler Thornblade:Building Lexicon for

Sentiment Analysis from Massive Collection of HTML Documents,Kaji, N. & Kitsuregawa, M., Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL)