ContextEnhanced Citation Sentiment Analysis Awais Athar Simone Teufel

  • Slides: 12
Download presentation
Context-Enhanced Citation Sentiment Analysis Awais Athar & Simone Teufel

Context-Enhanced Citation Sentiment Analysis Awais Athar & Simone Teufel

Sentiment Analysis of Citations

Sentiment Analysis of Citations

Challenges in Citation Sentiment Analysis • Negative sentiment is ‘politically dangerous’- (Ziman, 1968) •

Challenges in Citation Sentiment Analysis • Negative sentiment is ‘politically dangerous’- (Ziman, 1968) • Personal biases are hedged - (Hyland, 1995) • Criticism is ‘sweetened’ - (Mac. Roberts and Mac. Roberts, 1984; Hornsey et al. , 2008) “While SCL has been successfully applied to POS tagging and Sentiment Analysis (Blitzer et al. , 2006), its effectiveness for parsing was rather unexplored. ”

Problem: Context is Ignored

Problem: Context is Ignored

Our Contributions • A new citation sentiment corpus – contains citations annotated with the

Our Contributions • A new citation sentiment corpus – contains citations annotated with the dominant sentiment in the context – closer to the truth than single-sentence citations – increases citation sentiment coverage • Exploring effects of using context windows of different lengths on citation sentiment analysis

Corpus Construction • • Incoming citations to 20 papers 1, 741 citations (from >800

Corpus Construction • • Incoming citations to 20 papers 1, 741 citations (from >800 papers) Window length of 4 4 -class scheme – objective/neutral – positive – negative – e xcluded

Annotation Unit is the Citation • Problem – There may be more than 1

Annotation Unit is the Citation • Problem – There may be more than 1 sentiment /citation • Solution – For Gold Standard: assume last sentiment is what is really meant – For Automatic Treatment: merge citation context into one single sentence

Experiments • • SVM / 10 fold cross-validation Each citation as a feature set

Experiments • • SVM / 10 fold cross-validation Each citation as a feature set n-grams of length 1 to 3 Dependency triplets (Athar, 2010) det_results_The nsubj_good_results cop_good_were

Effect of Context Size

Effect of Context Size

Comparison with Athar (2010)

Comparison with Athar (2010)

Conclusion • Detection of citation sentiment in context around citation, not just citation sentence.

Conclusion • Detection of citation sentiment in context around citation, not just citation sentence. • New, large, context-aware citation corpus • Result F=0. 73 (sentiment is harder to find in science) • Improvement: Use coherence features to find variable window in each document A Athar and S Teufel, "Detection of implicit citations for sentiment detection", Accepted in Workshop on Detecting Structure in Scholarly Discourse 2012, ACL 2012.

Thank you! Questions?

Thank you! Questions?