Elamur Research⌕
☰
home / papers / 2008 / A Joint Model of Text and Aspect Ratings for Sentiment Summarization

A Joint Model of Text and Aspect Ratings for Sentiment Summarization

Summary

Titov and McDonald address a core step in aspect-based sentiment summarization: finding which pieces of a review talk about each rated aspect, such as service or location. Aspect labels at the sentence level are rarely available. Numeric per-aspect ratings, however, often come with online reviews, so the authors use them as weak supervision.

They propose the Multi-Aspect Sentiment model (MAS). It combines Multi-Grain LDA (MG-LDA), a topic model with document-level global topics and sliding-window local topics, with one maximum entropy rating predictor per aspect. Each predictor may only use words assigned to its own linked topic. This pushes that topic to capture text that actually explains the aspect's rating, and it makes the mapping from topics to aspects explicit.

The authors tested sentence labeling on 779 hand-annotated review sentences. MAS reaches average precision close to that of a supervised classifier, without any aspect-labeled text. The approach could generalize to any sequential data that comes with correlated signals, such as user tags or blog labels.

Key points

  • MAS has two parts:
    • an MG-LDA backbone, in which local topics tend to reflect ratable aspects and global topics reflect properties of particular items;
    • per-aspect MaxEnt classifiers, each tied to designated topics.
  • Ratings of different aspects are correlated, and some reviews express only general opinions. To handle this, an aspect's rating distribution is first estimated from the overall sentiment rating. The aspect's own topic words then supply corrections to that estimate.
  • The evaluation set has 779 hand-labeled sentences: 164 about service, 176 about location and 263 about rooms. The rest were irrelevant to the rated aspects.
  • MAS average precision:
    • 75.8% for service;
    • 85.5% for location;
    • 75.0%, 74.5%, 87.6% and 79.8% for rooms, using 1–4 topics per aspect.
  • A supervised MaxEnt baseline with unigram and bigram features, trained with 10-fold cross-validation, scores 80.8%, 94.0% and 88.3% on the same three aspects.

Interesting ideas

  • I1. Use existing metadata as weak labels. Star ratings, tags and similar user-supplied signals already in the data can steer an unsupervised model toward meaningful structure. This avoids annotation cost in any domain where such signals are common.
  • I2. Restrict a predictor to one latent component's output. A predictor that may only see words from its designated topic forces that topic to become interpretable and tied to a named label. This is a general way to anchor latent variables to known categories.
  • I3. Model the residual, not the whole signal. A shared baseline predicts the overall level, here from the overall rating, and each component only explains deviations from it. This keeps generic or correlated cues from taking over the specific components, which is useful whenever target signals are strongly correlated.
  • I4. Overlapping sliding windows for local context. Overlapping windows over sentences let a topic model capture local themes without expensive transition modeling. This is a cheap trick for segmentation-like tasks.
  • I5. Benchmark against a supervised upper bound. Comparing the weakly supervised model to a supervised classifier shows how close it gets. This is a convenient evaluation design when no comparable unsupervised baseline can map outputs to labels.

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

108 papers of our library cite it (of the 140884 whose bibliographies we read); 11 of them have a page here, 97 not yet.

Cite

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

@inproceedings{titov2008joint,
  title = {{A Joint Model of Text and Aspect Ratings for Sentiment Summarization}},
  author = {Titov, Ivan and McDonald, Ryan},
  booktitle = {Proceedings of ACL-08: HLT},
  year = {2008},
  url = {https://aclanthology.org/P08-1036/}
}

References

23 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David M. Blei and Jon D. McAuliffe. 2008. Supervised topic models. In Advances in Neural Information Processing Systems (NIPS). find
  2. D.M. Blei, A.Y. Ng, and M.I. Jordan. 2003. Latent Dirichlet allocation. Journal of Machine Learning Research, 3(5):993–1022. find our card
  3. S.R.K. Branavan, H. Chen, J. Eisenstein, and R. Barzilay. 2008. Learning document-level semantic properties from free-text annotations. In Proceedings of the Annual Conference of the Association for Computational Linguistics. find
  4. G. Carenini, R. Ng, and A. Pauls. 2006. Multi-Document Summarization of Evaluative Text. In Proceedings of the Conference of the European Chapter of the Association for Computational Linguistics. find
  5. M. Gamon, A. Aue, S. Corston-Oliver, and E. Ringger. 2005. Pulse: Mining customer opinions from free text. In Proc. of the 6th International Symposium on Intelligent Data Analysis, pages 121–132. find
  6. S. Geman and D. Geman. 1984. Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 6:721–741. find
  7. T. L. Griffiths and M. Steyvers. 2004. Finding scientific topics. Proceedings of the Natural Academy of Sciences, 101 Suppl 1:5228–5235. find
  8. T. L. Griffiths, M. Steyvers, D. M. Blei, and J. B. Tenenbaum. 2004. Integrating topics and syntax. In Advances in Neural Information Processing Systems. find
  9. A. Gruber, Y. Weiss, and M. Rosen-Zvi. 2007. Hidden Topic Markov Models. In Proceedings of the Conference on Artificial Intelligence and Statistics. find
  10. M. Hu and B. Liu. 2004a. Mining and summarizing customer reviews. In Proceedings of the 2004 ACM SIGKDD international conference on Knowledge discovery and data mining, pages 168–177. ACM Press New York, NY, USA. find
  11. M. Hu and B. Liu. 2004b. Mining Opinion Features in Customer Reviews. In Proceedings of Nineteenth National Conference on Artificial Intellgience. find
  12. C. Manning and M. Schutze. 1999. Foundations of Statistical Natural Language Processing. MIT Press. find
  13. Q. Mei, X. Ling, M. Wondra, H. Su, and C.X. Zhai. 2007. Topic sentiment mixture: modeling facets and opinions in weblogs. In Proceedings of the 16th International Conference on World Wide Web, pages 171–180. find
  14. Radford Neal. 1992. Connectionist learning of belief networks. Artificial Intelligence, 56:71–113. find
  15. B. Pang, L. Lee, and S. Vaithyanathan. 2002. Thumbs up? Sentiment classification using machine learning techniques. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. find our card
  16. A.M. Popescu and O. Etzioni. 2005. Extracting product features and opinions from reviews. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). find our card
  17. B. Snyder and R. Barzilay. 2007. Multiple Aspect Ranking using the Good Grief Algorithm. In Proceedings of the Joint Conference of the North American Chapter of the Association for Computational Linguistics and Human Language Technologies, pages 300–307. find
  18. I. Titov and R. McDonald. 2008. Modeling online reviews with multi-grain topic models. In Proceedings of the 17h International Conference on World Wide Web. find
  19. P. Turney. 2002. Thumbs up or thumbs down? Sentiment orientation applied to unsupervised classification of reviews. In Proceedings of the Annual Conference of the Association for Computational Linguistics. find
  20. Hanna M. Wallach. 2006. Topic modeling; beyond bag of words. In International Conference on Machine Learning. find
  21. Xuerui Wang and Andrew McCallum. 2005. A note on topical n-grams. Technical Report UM-CS-2005-071, University of Massachusetts. find
  22. J. Wiebe. 2000. Learning subjective adjectives from corpora. In Proceedings of the National Conference on Artificial Intelligence. find
  23. L. Zhuang, F. Jing, and X.Y. Zhu. 2006. Movie review mining and summarization. In Proceedings of the 15th ACM international conference on Information and knowledge management (CIKM), pages 43–50. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/titov-2008-joint-model-text-aspect-ratings/