Elamur Research⌕
☰
home / papers / 2013 / Joint Inference for Fine-grained Opinion Extraction

Joint Inference for Fine-grained Opinion Extraction

Paper · 2013 · Bishan Yang and Claire Cardie

Summary

The paper tackles fine-grained opinion extraction on news text. The task has two parts: finding spans of opinion expressions, opinion targets and opinion holders, and linking them through IS-ABOUT and IS-FROM relations. Earlier systems mostly ran these steps as a pipeline, so mistakes in one stage carried into the next. Yang and Cardie instead split the problem into local predictors and combine them in a single integer linear program (ILP):

  • a CRF sequence tagger supplies span-level probabilities for each entity type;
  • logistic regression classifiers score opinion–argument pairs, and also score the option that an opinion has no explicit argument (an implicit holder or target).

Constraints in the ILP keep the entity labels and relations globally consistent. Experiments on the MPQA corpus show significant gains over pipeline methods and over baselines that solve each subtask alone. The improvement comes mainly from higher recall. This matters because it shows that cheap, separately trained components can be merged into a better joint decision without training one large joint model.

Key points

  • Entity tagging: opinion entities are tagged with a CRF using IOB encoding over four labels: D (opinion expression), T (target), H (holder) and N (none). Span probabilities are computed with forward-backward.
  • Candidate generation:
    • The training data is split into 10 parts, and 50-best CRF outputs are collected for each part. Using more n-best outputs did not help.
    • Pairs are also generated from the most frequent shortest dependency-path patterns, such as dobj and ccomp paths for IS-ABOUT, and nsubj and poss paths for IS-FROM.
  • Ablations: three variants each perform worse than the full model:
    • ILP-W/O-ENTITY, which drops the entity potentials and performs poorly;
    • ILP-W-SINGLE-RE, which models only one relation type;
    • ILP-W/O-IMPLICIT-RE, which ignores implicit arguments.
  • Comparison with Choi et al. (2006): the new ILP formulation beats Choi et al.'s ILP on holders, expressions and IS-FROM, while using fewer variables and only one parameter.
  • Speed and error sources: GLPK's branch-and-cut solver found optimal solutions for 1000 sentences in 0.2 seconds. The CRF baselines output only about 60% as many entities as the gold standard. Many remaining errors come from correct answers that never appeared among the candidates.

Interesting ideas

  • I1. Use a tagger's marginals as ILP scores. The probability that a span gets a given label can serve directly as a score inside an optimizer. Any existing CRF tagger could be plugged into a constrained global decoder this way, without retraining.
  • I2. Make "no argument" an explicit choice. Scoring the empty-argument option alongside real candidates lets the model decide that an argument is simply absent. This could carry over to dropped subjects, implicit arguments in semantic role labeling, or missing slots in information extraction.
  • I3. Score relations as a probability difference. Defining the relation score as P(positive) minus P(negative) yields signed weights. A link is then selected only when the classifier leans towards it, which avoids hand-tuned thresholds.
  • I4. Build candidates from two sources. Combining n-best tagger outputs with frequent dependency-path patterns gives good coverage while limiting the number of candidates. The authors' error analysis also suggests checking candidate recall as its own diagnostic, since joint inference cannot recover answers that were never proposed.
  • I5. Ablate by removing ILP components. Taking out one potential at a time (entities, the second relation type, implicit links) shows how much each information source contributes. This evaluation design is easy to reuse for other joint models.

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

 Institutions

🎓Cornell UniversityIthaca · United States89 works · 58,422 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

Cite

Cite as Bishan Yang and Claire Cardie. 2013. Joint Inference for Fine-grained Opinion Extraction. Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

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

@inproceedings{yang2013joint,
  title = {{Joint Inference for Fine-grained Opinion Extraction}},
  author = {Bishan Yang and Claire Cardie},
  booktitle = {Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2013},
  url = {https://aclanthology.org/P13-1161/}
}

References

30 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. E. Breck, Y. Choi, and C. Cardie. 2007. Identifying expressions of opinion in context. In Proceedings of the 20th international joint conference on Artifical intelligence, pages 2683–2688. Morgan Kaufmann Publishers Inc. find
  2. Yejin Choi, Claire Cardie, Ellen Riloff, and Siddharth Patwardhan. 2005. Identifying sources of opinions with conditional random fields and extraction patterns. In Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing, pages 355–362. Associat…
  3. Y. Choi, E. Breck, and C. Cardie. 2006. Joint extraction of entities and relations for opinion recognition. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing, pages 431–439. Association for Computational Linguistics. find
  4. D. Das, A.F.T. Martins, and N.A. Smith. 2012. An exact dual decomposition algorithm for shallow semantic parsing with constraints. Proceedings of* SEM.[ii, 10, 50]. find
  5. Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang-Rui Wang, and Chih-Jen Lin. 2008. Liblinear: A library for large linear classification. The Journal of Machine Learning Research, 9:1871–1874. find
  6. M. Hu and B. Liu. 2004. Mining opinion features in customer reviews. In Proceedings of the National Conference on Artificial Intelligence, pages 755–760. find
  7. Menlo Park, CA; Cambridge, MA; London; AAAI Press; MIT Press; 1999. find
  8. Richard Johansson and Alessandro Moschitti. 2010a. Reranking models in fine-grained opinion analysis. In Proceedings of the 23rd International Conference on Computational Linguistics, pages 519–527. Association for Computational Linguistics. find
  9. Richard Johansson and Alessandro Moschitti. 2010b. Syntactic and semantic structure for opinion expression detection. In Proceedings of the Fourteenth Conference on Computational Natural Language Learning, pages 67–76. Association for Computational Linguistics. find
  10. Richard Johansson and Alessandro Moschitti. 2012. Relational features in fine-grained opinion analysis. find
  11. Soo-Min Kim and Eduard Hovy. 2006. Extracting opinions, opinion holders, and topics expressed in online news media text. In Proceedings of the Workshop on Sentiment and Subjectivity in Text, pages 1– 8. Association for Computational Linguistics.
  12. N. Kobayashi, K. Inui, and Y. Matsumoto. 2007. Extracting aspect-evaluation and aspect-of relations in opinion mining. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL), pages 1065– 1074. find
  13. John Lafferty, Andrew McCallum, and Fernando CN Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. find
  14. K. Liu, L. Xu, and J. Zhao. 2012. Opinion target extraction using word-based translation model. In Proceedings of the conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics. find
  15. Bing Liu. 2012. Sentiment analysis and opinion mining. Synthesis Lectures on Human Language Technologies, 5(1):1–167. find
  16. Bo Pang and Lillian Lee. 2008. Opinion mining and sentiment analysis. Now Pub. find
  17. V. Punyakanok, D. Roth, and W. Yih. 2008. The importance of syntactic parsing and inference in semantic role labeling. Computational Linguistics, 34(2):257–287.
  18. Guang Qiu, Bing Liu, Jiajun Bu, and Chun Chen. 2009. Expanding domain sentiment lexicon through double propagation. In Proceedings of the 21st international jont conference on Artifical intelligence, pages 1199–1204. Morgan Kaufmann Publishers Inc. find
  19. G. Qiu, B. Liu, J. Bu, and C. Chen. 2011. Opinion word expansion and target extraction through double propagation. Computational linguistics, 37(1):9– 27.
  20. R. Quirk, S. Greenbaum, G. Leech, J. Svartvik, and D. Crystal. 1985. A comprehensive grammar of the English language, volume 397. Cambridge Univ Press. find
  21. D. Roth and W. Yih. 2004. A linear programming formulation for global inference in natural language tasks. Defense Technical Information Center.
  22. J. Ruppenhofer, S. Somasundaran, and J. Wiebe. 2008. Finding the sources and targets of subjective expressions. In Proceedings of LREC. find
  23. Vivek Srikumar and Dan Roth. 2011. A joint model for extended semantic role labeling. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 129–139. Association for Computational Linguistics. find
  24. V. Stoyanov and C. Cardie. 2008. Topic identification for fine-grained opinion analysis. In Proceedings of the 22nd International Conference on Computational Linguistics-Volume 1, pages 817–824. Association for Computational Linguistics. find
  25. J. Wiebe, T. Wilson, and C. Cardie. 2005. Annotating expressions of opinions and emotions in language. Language Resources and Evaluation, 39(2):165– 210. find
  26. Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. 2009. Recognizing contextual polarity: An exploration of features for phrase-level sentiment analysis. Computational linguistics, 35(3):399–433.
  27. Theresa Wilson. 2008. Fine-Grained Subjectivity Analysis. Ph.D. thesis, Ph. D. thesis, University of Pittsburgh. Intelligent Systems Program. find
  28. Y. Wu, Q. Zhang, X. Huang, and L. Wu. 2009. Phrase dependency parsing for opinion mining. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 3-Volume 3, pages 1533–1541. Association for Computational Linguistics.
  29. B. Yang and C. Cardie. 2012. Extracting opinion expressions with semi-markov conditional random fields. In Proceedings of the conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics. find
  30. Lei Zhang, Bing Liu, Suk Hwan Lim, and Eamonn O’Brien-Strain. 2010. Extracting and ranking product features in opinion documents. In Proceedings of the 23rd International Conference on Computational Linguistics: Posters, pages 1462–1470. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2013-joint-inference-fine-grained-opinion/