Supervised and Unsupervised Methods in Employing Discourse Relations for Improving Opinion Polarity Classification

Summary

The paper asks whether a discourse-level annotation scheme for opinions can be turned into working algorithms that improve polarity classification (positive, negative, neutral). The authors use the scheme of Somasundaran et al. (2008), annotated over the AMI meeting corpus. In that scheme, opinion targets are linked by same or alternative relations, and opinions are linked by reinforcing or non-reinforcing frame relations. They move these annotations onto Dialog Act (DA) units and train a supervised local classifier to give initial labels. Two global inference methods then build on those labels. The first is a supervised ICA variant, which adds relational features derived from the polarities of discourse neighbours. The second is an unsupervised ILP formulation, which stays close to the local predictions while respecting discourse constraints. Both methods are significantly more accurate than the local-only baseline. They improve in different ways, and a simple hybrid of the two outperforms each one. The gains come mainly from much higher recall on polar classes with precision intact. This is early empirical evidence that discourse relations can be exploited computationally for sentiment analysis, even when the starting labels are noisy.

Key points

Data: The AMI meeting corpus, made up of scenario-based design meetings about a TV remote. Its provided DA segmentation defines the classification units. When a DA contains several opinions with conflicting polarities, one label is picked at random.
Local baseline: Features come from polarity lexicons (as in Wilson et al., 2005), DA tags, and unigrams. Because only the Local stage is supervised, the end-to-end ILP pipeline counts as semi-supervised.
ICA: Follows Lu and Getoor (2003), with ten groups of relational features. These include the share of neighbours with each polarity under each frame or target relation, the same-speaker variants, and the majority neighbour polarity.
ILP: Encodes constraints such as equal polarity for same+reinforcing links and opposite polarity for alternative+reinforcing links. In worked examples, a single confident correct prediction spreads through these links and fixes errors in nodes that are only indirectly connected.
Future work: All results use gold-standard discourse relations. The authors name automatic recognition of those relations as the next step.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Discourse analysis

Cited by

38 papers of our library cite it (of the 140673 whose bibliographies we read); 13 of them have a page here, 25 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{somasundaran2009supervised,
  title = {{Supervised and Unsupervised Methods in Employing Discourse Relations for Improving Opinion Polarity Classification}},
  author = {Somasundaran, Swapna and Namata, Galileo and Wiebe, Janyce and Getoor, Lise},
  booktitle = {Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing},
  year = {2009},
  url = {https://aclanthology.org/D09-1018/}
}

References

31 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

N. Asher, F. Benamara, and Y. Mathieu. 2008. Distilling opinion in discourse: A preliminary study. COLING-2008. find
M. Bansal, C. Cardie, and L. Lee. 2008. The power of negative thinking: Exploiting label disagreement in the min-cut classification framework. In COLING-2008. find
M. Bilgic, G. M. Namata, and L. Getoor. 2007. Combining collective classification and link prediction. find
In Workshop on Mining Graphs and Complex Structures at the IEEE International Conference on Data Mining. find
J. Carletta, S. Ashby, S. Bourban, M. Flynn, M. Guillemot, T. Hain, J. Kadlec, V. Karaiskos, W. Kraaij, M. Kronenthal, G. Lathoud, M. Lincoln, A. Lisowska, I. McCowan, W. Post, D. Reidsma, and P. Wellner. 2005. The ami meetings corpus. In Proceedings of the Measuring Behavior Symposium on ”Annotatin… find
Y. Choi, E. Breck, and C. Cardie. 2006. Joint extraction of entities and relations for opinion recognition. In EMNLP 2006. find
P. Denis and J. Baldridge. 2007. Joint determination of anaphoricity and coreference resolution using integer programming. In HLT-NAACL 2007. find our card
A. Devitt and K. Ahmad. 2007. Sentiment polarity identification in financial news: A cohesion-based approach. In ACL 2007. find
A. B. Goldberg and X. Zhu. 2006. Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization. In HLT-NAACL 2006 Workshop on Textgraphs: Graph-based Algorithms for Natural Language Processing. find
A. Haghighi, K. Toutanova, and C. Manning. 2005. A joint model for semantic role labeling. In CoNLL. find
H. Kanayama and T. Nasukawa. 2006. Fully automatic lexicon expansion for domain-oriented sentiment analysis. In EMNLP-2006, pages 355–363, Sydney, Australia. find
A. Kennedy and D. Inkpen. 2006. Sentiment classification of movie reviews using contextual valence shifters. Computational Intelligence, 22(2):110– 125. find
Q. Lu and L. Getoor. 2003. Link-based classification. In Proceedings of the International Conference on Machine Learning (ICML). find
A. Moschitti, D. Pighin, and R. Basili. 2006. Semantic role labeling via tree kernel joint inference. In CoNLL. find
A. Moschitti. 2009. Syntactic and semantic kernels for short text pair categorization. In EACL. find
D. Neiberg, K. Elenius, and K. Laskowski. 2006. Emotion recognition in spontaneous speech using gmms. In INTERSPEECH 2006 ICSLP. find
J. Neville and D. Jensen. 2000. Iterative classification in relational data. In In Proc. AAAI-2000 Workshop on Learning Statistical Models from Relational Data, pages 13–20. AAAI Press. find
B. Pang and L. Lee. 2004. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts. In ACl 2004. find
L. Polanyi and A. Zaenen, 2006. Contextual Valence Shifters. Computing Attitude and Affect in Text: Theory and Applications. find
A.-M. Popescu and O. Etzioni. 2005. Extracting product features and opinions from reviews. In HLT-EMNLP 2005. find
S. Raaijmakers, K. Truong, and T. Wilson. 2008. Multimodal subjectivity analysis of multiparty conversation. In EMNLP. find
M. Richardson and P. Domingos. 2006. Markov logic networks. Mach. Learn., 62(1-2):107–136. find
D. Roth and W. Yih. 2004. A linear programming formulation for global inference in natural language tasks. In Proceedings of CoNLL-2004, pages 1–8. Boston, MA, USA. find our card
K. Sadamitsu, S. Sekine, and M. Yamamoto. 2008. Sentiment analysis based on probabilistic models using inter-sentence information. In LREC’08. find
S. Somasundaran, J. Ruppenhofer, and J. Wiebe. 2007. Detecting arguing and sentiment in meetings. In SIGdial Workshop on Discourse and Dialogue 2007. find
S. Somasundaran, J. Wiebe, and J. Ruppenhofer. 2008. Discourse level opinion interpretation. In Coling 2008. find
H. Takamura, T. Inui, and M. Okumura. 2007. Extracting semantic orientations of phrases from dictionary. In HLT-NAACL 2007. find
B. Taskar, M. Wong, P. Abbeel, and D. Koller. 2004. Link prediction in relational data. In Neural Information Processing Systems. find
M. Thomas, B. Pang, and L. Lee. 2006. Get out the vote: Determining support or opposition from congressional floor-debate transcripts. In EMNLP 2006. find
T. Wilson, J. Wiebe, and P. Hoffmann. 2005. Recognizing contextual polarity in phrase-level sentiment analysis. In HLT-EMNLP 2005. find
I. H. Witten and E. Frank. 2002. Data mining: practical machine learning tools and techniques with java implementations. SIGMOD Rec., 31(1):76–77. find

↑ back to top