Elamur Research⌕
☰
home / papers / 2010 / An Unsupervised Aspect-Sentiment Model for Online Reviews

An Unsupervised Aspect-Sentiment Model for Online Reviews

Summary

Review sites cover so many kinds of products and services that annotated training data rarely exists for each one. The authors propose a fully unsupervised pipeline that finds the aspects discussed in reviews and estimates the polarity of adjectives for each aspect separately. For aspects, they run standard LDA (GibbsLDA++) on individual sentences rather than whole reviews and keep the topic count small, so topics map directly to rateable aspects. They pick the number of topics with a cluster-consistency validation procedure. For sentiment, they build positive and negative seed adjectives automatically from morphological negation cues, then propagate polarity within each aspect. The main data is the Citysearch New York restaurant corpus of Ganu et al. (2009), with Amazon netbook reviews added to test a second domain. The automatic seeds rank adjectives somewhat less closely to human judgments than a manual seed list, but the results remain usable. The system matches more complex semi-supervised, knowledge-heavy methods without needing lexicons. This matters for new domains, noisy text and other languages.

Key points

  • Data:
    • More than 50,000 Citysearch New York restaurant reviews (Ganu et al., 2009).
    • 1,086 Amazon reviews of four netbooks.
    • A gold subset of 3,400 sentences labeled with six aspect classes (Food & Drink, Service, Price, Atmosphere, Anecdotes, Miscellaneous) and four sentiment values (Positive, Negative, Neutral, Conflict).
  • Local LDA setup: out-of-the-box settings (α = 0.1, β = 0.1, 3000 iterations) with no tuning. The model was tried with 10–20 aspects, and the order was selected using a consistency function following Niu et al. (2007).
  • Sentiment evaluation: adjective rankings were compared to annotator judgments using Kendall's τ and Kendall's distance over eight inferred aspects.
    • Automatic seeds averaged τ = 0.36 and D = 0.32.
    • The manual seed set of Fahrni and Klenner (128 positive, 88 negative adjectives) averaged τ = 0.45 and D = 0.27.
  • Weakest aspects: agreement was lowest for Main Dishes and Food – General. These aspects had the largest share of adjectives that annotators considered neutral, such as cuisine names.
  • Better coverage than the manual scheme: the inferred aspects surfaced categories that the manual labels missed or merged, for example a distinct Drinks aspect and a Service aspect separate from Staff.

Interesting ideas

  • I1. Treat each sentence as a document in a topic model. Shrinking the context pushes LDA toward fine-grained, local themes instead of document-wide subjects. This is a cheap trick for any short-text or aspect-level clustering task.
  • I2. Choose the number of clusters by stability. Compare how consistently pairs of items end up grouped together across different model orders. This gives a principled, label-free way to set k in other unsupervised settings.
  • I3. Derive sentiment seeds from negation morphology. Word pairs formed with negating affixes yield opposite-polarity, domain-relevant seeds without a hand-built lexicon. The authors suggest the idea could carry over to other languages, which could suit morphologically rich languages such as Ukrainian.
  • I4. Compute polarity per aspect, not globally. The same adjective can flip polarity depending on the aspect: "warm" is positive for staff but slightly negative for food, and "cheap" behaves differently for food than for decor. Aspect-conditioned lexicons capture this context-dependent sentiment.
  • I5. Rank "neutral" words within each aspect to reveal hidden associations. Ranking objective terms, such as cuisine types, can expose implicit sentiment that annotators overlook. This is a useful exploratory analysis tool for any review corpus.

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

Cited by

74 papers of our library cite it (of the 140884 whose bibliographies we read); 3 of them have a page here, 71 not yet.

Cite

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

@inproceedings{brody2010unsupervised,
  title = {{An Unsupervised Aspect-Sentiment Model for Online Reviews}},
  author = {Brody, Samuel and Elhadad, Noemie},
  booktitle = {Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2010},
  url = {https://aclanthology.org/N10-1122/}
}

References

24 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Benamara, Farah, Carmine Cesarano, Antonio Picariello, Diego Reforgiato, and V. S. Subrahmanian. 2007. Sentiment analysis: Adjectives and adverbs are better than adjectives alone. In Proc. of the International Conference on Weblogs and Social Media (ICWSM). find
  2. Blei, David M., Andrew Y. Ng, and Michael I. Jordan. 2003. Latent dirichlet allocation. Journal of Machine Learning Research 3:993–1022. find our card
  3. Blitzer, John, Mark Dredze, and Fernando Pereira. 2007. Biographies, bollywood, boom-boxes and blenders: Domain adaptation for sentiment classification. In Proc. of the 45th Annual Meeting of the Association of Computational Linguistics. ACL, Prague, Czech Republic, pages 440–447. find our card
  4. Briscoe, Ted and John Carroll. 2002. Robust accurate statistical annotation of general text. In Proc. of the 3rd LREC. Las Palmas, Gran Canaria, pages 1499–1504. find our card
  5. Fahrni, Angela and Manfred Klenner. 2008. Old Wine or Warm Beer: Target-Specific Sentiment Analysis of Adjectives. In Proc.of the Symposium on Affective Language in Human and Machine, AISB 2008 Convention. pages 60 – 63. find
  6. Ganu, Gayatree, Noemie Elhadad, and Amelie Marian. 2009. Beyond the stars: Improving rating predictions using review text content. In WebDB. find
  7. Hatzivassiloglou, Vasileios and Kathleen R. McKeown. 1997. Predicting the semantic orientation of adjectives. In Proc. of the 35th Annual Meeting of the Association for Computational Linguistics. ACL, Madrid, Spain, pages 174–181. find
  8. Hu, Minqing and Bing Liu. 2004. Mining and summarizing customer reviews. In KDD ’04: Proc. of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, New York, NY, USA, pages 168–177. find
  9. Jijkoun, Valentin and Katja Hofmann. 2009. Generating a non-english subjectivity lexicon: Relations that matter. In Proc. of the 12th Conference of the European Chapter of the ACL (EACL 2009). ACL, Athens, Greece, pages 398–405. find
  10. Lange, Tilman, Volker Roth, Mikio L. Braun, and Joachim M. Buhmann. 2004. Stability-based validation of clustering solutions. Neural Comput. 16(6):1299–1323. find
  11. Lerman, Kevin, Sasha Blair-Goldensohn, and Ryan Mc-Donald. 2009. Sentiment summarization: evaluating and learning user preferences. In EACL ’09: Proc. of the 12th Conference of the European Chapter of the Association for Computational Linguistics. ACL, Morristown, NJ, USA, pages 514–522. find
  12. Levine, Erel and Eytan Domany. 2001. Resampling method for unsupervised estimation of cluster validity. Neural Comput. 13(11):2573–2593. find
  13. Mei, Qiaozhu, Xu Ling, Matthew Wondra, Hang Su, and ChengXiang Zhai. 2007. Topic sentiment mixture: modeling facets and opinions in weblogs. In WWW ’07: Proc. of the 16th international conference on World Wide Web. ACM, New York, NY, USA, pages 171–180. find
  14. Niu, Zheng-Yu, Dong-Hong Ji, and Chew-Lim Tan. 2007. I2r: three systems for word sense discrimination, chinese word sense disambiguation, and english word sense disambiguation. In SemEval ’07: Proc. of the 4th International Workshop on Semantic Evaluations. ACL, Morristown, NJ, USA, pages 177–182. find
  15. Pang, Bo and Lillian Lee. 2005. Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales. In Proc. of the ACL. pages 115–124. find our card
  16. Pang, Bo and Lillian Lee. 2008. Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval 2(1-2):1–135. find
  17. Pang, Bo, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up?: sentiment classification using machine learning techniques. In EMNLP ’02: Proc. of the conference on Empirical methods in natural language processing. ACL, Morristown, NJ, USA, pages 79–86. find our card
  18. Popescu, Ana-Maria and Oren Etzioni. 2005. Extracting product features and opinions from reviews. In HLT ’05: Proc. of the conference on Human Language Technology and Empirical Methods in Natural Language Processing. ACL, Morristown, NJ, USA, pages 339–346. find our card
  19. Snyder, Benjamin and Regina Barzilay. 2007. Multiple aspect ranking using the good grief algorithm. In Candace L. Sidner, Tanja Schultz, Matthew Stone, and ChengXiang Zhai, editors, HLT-NAACL. The Association for Computational Linguistics, pages 300–307. find
  20. Titov, Ivan and Ryan McDonald. 2008a. A joint model of text and aspect ratings for sentiment summarization. In Proc. of ACL-08: HLT. ACL, Columbus, Ohio, pages 308–316. find our card
  21. Titov, Ivan and Ryan McDonald. 2008b. Modeling online reviews with multi-grain topic models. In WWW ’08: Proc. of the 17th international conference on World Wide Web. ACM, New York, NY, pages 111–120. find
  22. Turney, Peter. 2002. Thumbs up or thumbs down? semantic orientation applied to unsupervised classification of reviews. In Proc. of 40th Annual Meeting of the Association for Computational Linguistics. ACL, Philadelphia, Pennsylvania, USA, pages 417–424. find our card
  23. Zagibalov, Taras and John Carroll. 2008. Automatic seed word selection for unsupervised sentiment classification of chinese text. In COLING ’08: Proc. of the 22nd International Conference on Computational Linguistics. ACL, Morristown, NJ, USA, pages 1073–1080. find
  24. Zhu, X. and Z. Ghahramani. 2002. Learning from labeled and unlabeled data with label propagation. Technical report, CMU-CALD-02. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/brody-2010-unsupervised-aspect-sentiment-model-online/