Elamur Research⌕
☰
home / papers / 2009 / Phrase Dependency Parsing for Opinion Mining

Phrase Dependency Parsing for Opinion Mining

Summary

The paper tackles opinion mining from product reviews. An opinion unit is defined as a product feature, an opinion expression and a polarity, and the core subtask is deciding which features and expressions are actually linked. The authors note that product features are often multi-word phrases (e.g. "LCD screen"), which word-level dependency trees handle poorly. They therefore propose phrase dependency parsing: the output of a shallow parser (chunker) is merged with a lexical dependency tree, so each chunk becomes one node and its internal word dependencies are kept as a nested structure. Candidate features are NP/VP chunks, filtered by a domain language model trained on unlabeled reviews. Candidate opinion expressions come from a dictionary. Relations are then classified with an SVM that uses a new tree kernel, which treats dependencies inside phrases separately from those between phrases. Across five product domains, the phrase-tree kernel (SVM-PTree) beats the word-tree kernel (SVM-WTree) by 4.8% on average. This shows that phrase-level syntax helps relation extraction.

Key points

  • Pipeline in three stages: build the phrase dependency tree from chunking and dependency parsing, extract candidate features and opinion expressions, then classify feature–expression relations.
  • In the labeled corpus, more than 98% of product features fall within one NP or VP chunk. All NPs and VPs are therefore taken as candidates, while PPs and ADJPs are excluded.
  • Opinion expression candidates rely on a dictionary, motivated by Kobayashi et al. (2007). Their 5,550 expressions showed high coverage across domains.
  • Compared methods: an Adjacent baseline, SVM-1, SVM-2, SVM-WTree and SVM-PTree.
    • Learning methods beat Adjacent only in the three larger domains.
    • On the two small domains (e.g. Diaper, DVD Player), the simple Adjacent baseline did better.
  • Cross-domain training improved SVM-1, SVM-WTree and SVM-PTree by 4.6%, 8.6% and 10.3% on average. SVM-2, which uses domain-dependent features, dropped.

Interesting ideas

  • I1. Collapsing chunks into dependency-tree nodes. Chunks become single nodes while their internal word arcs are kept as nested substructure. This yields a compact tree that still holds long-distance links, which helps any tree-kernel or graph-based relation extractor working with multi-word entities.
  • I2. Resolving parser conflicts by trusting the dependency parser. When the shallow parser and the dependency parser disagree, the dependency parser wins and inconsistent chunks are dropped. The output is guaranteed to remain a valid tree, a cheap and robust rule for combining two imperfect analyzers.
  • I3. Separating local and global structure in the kernel. Intra-phrase dependencies are kept from dominating the comparison of overall tree shape. The same split could suit morphologically rich languages, where much structure sits inside phrases.
  • I4. Pruning high-recall candidates with a domain language model. Candidates are scored by an LM trained on cheap unlabeled in-domain text, which reduces noise before classification without extra annotation.
  • I5. Oracle ablations for pipeline diagnosis. The OERight and PFRight runs feed gold opinion expressions or gold features into relation extraction. They showed that opinion-expression errors hurt most, a useful way to decide which pipeline stage to improve first.

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: Natural language processing

Topics: Dependency parsing · Sentiment & emotion

Institutions

🎓Fudan UniversityShanghai · China45 works · 16 344 citations

Cite

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

@inproceedings{wu2009phrase,
  title = {{Phrase Dependency Parsing for Opinion Mining}},
  author = {Wu, Yuanbin and Zhang, Qi and Huang, Xuanjing and Wu, Lide},
  booktitle = {Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing},
  year = {2009},
  doi = {10.3115/1699648.1699700},
  url = {https://doi.org/10.3115/1699648.1699700}
}

References

23 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eric Breck, Yejin Choi, and Claire Cardie. 2007. Identifying expressions of opinion in context. In Proceedings of IJCAI-2007. 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 HLT/EMNLP. find
  3. Yejin Choi, Eric Breck, and Claire Cardie. 2006. Joint extraction of entities and relations for opinion recognition. In Proceedings EMNLP. find
  4. Aron Culotta and Jeffrey Sorensen. 2004. Dependency tree kernels for relation extraction. In In Proceedings of ACL 2004. find our card
  5. Kushal Dave, Steve Lawrence, and David M. Pennock. 2003. Mining the peanut gallery: opinion extraction and semantic classification of product reviews. In Proceedings of WWW 2003. find
  6. Minqing Hu and Bing Liu. 2004. Mining and summarizing customer reviews. In Proceedings of the ACM SIGKDD 2004. find
  7. Nitin Jindal and Bing Liu. 2008. Opinion spam and analysis. In Proceedings of WSDM ’08. find
  8. Thorsten Joachims, Nello Cristianini, and John Shawe-Taylor. 2001. Composite kernels for hypertext categorisation. In Proceedings of ICML ’01. find
  9. Soo-Min Kim and Eduard Hovy. 2004. Determining the sentiment of opinions. In Proceedings of Coling 2004. COLING. find our card
  10. Dan Klein and Christopher D. Manning. 2002. Fast exact inference with a factored model for natural language parsing. In In Advances in Neural Information Processing Systems. find
  11. Nozomi Kobayashi, Kentaro Inui, and Yuji Matsumoto. 2007. Extracting aspect-evaluation and aspect-of relations in opinion mining. In Proceedings of EMNLP-CoNLL 2007. find
  12. Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up? Sentiment classification using machine learning techniques. In Proc. of EMNLP 2002. find our card
  13. Ana-Maria Popescu and Oren Etzioni. 2005. Extracting product features and opinions from reviews. In Proceedings of HLT/EMNLP. find our card
  14. E. Riloff and W. Phillips. 2004. An introduction to the sundance and autoslog systems. In University of Utah School of Computing Technical Report UUCS-04-015. find
  15. Beatrice Santorini and Anthony Kroch. 2007. The syntax of natural language: An online introduction using the Trees program. http://www.ling.upenn.edu/beatrice/syntax-textbook. ↗
  16. Swapna Somasundaran, Janyce Wiebe, and Josef Ruppenhofer. 2008. Discourse level opinion interpretation. In Proceedings of COLING 2008. find
  17. Hiroya Takamura, Takashi Inui, and Manabu Okumura. 2005. Extracting semantic orientations of words using spin model. In Proceedings of ACL’05. find
  18. L. Tesnière. 1959. Eléments de syntaxe structurale. Editions Klincksieck. find
  19. Janyce Wiebe, Theresa Wilson, and Claire Cardie. 2005. Annotating expressions of opinions and emotions in language. Language Resources and Evaluation, 39(2/3). find
  20. Theresa Wilson, Paul Hoffmann, Swapna Somasundaran, Jason Kessler, Janyce Wiebe, Yejin Choi, Claire Cardie, Ellen Riloff, and Siddharth Patwardhan. 2005a. Opinionfinder: A system for subjectivity analysis. In Demonstration Description in Conference on Empirical Methods in Natural Language Processing… find
  21. Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. 2005b. Recognizing contextual polarity in phrase-level sentiment analysis. In Proceedings of HLT-EMNLP. find our card
  22. Fei Xia and Martha Palmer. 2001. Converting dependency structures to phrase structures. In HLT ’01: Proceedings of the first international conference on Human language technology research. find
  23. Qi Zhang, Yuanbin Wu, Tao Li, Mitsunori Ogihara, Joseph Johnson, and Xuanjing Huang. 2009. Mining product reviews based on shallow dependency parsing. In Proceedings of SIGIR 2009. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2009-phrase-dependency-parsing-opinion-mining/