Elamur Research⌕
☰
home / papers / 2018 / Text Segmentation as a Supervised Learning Task

Text Segmentation as a Supervised Learning Task

Paper · 2018 · Omri Koshorek, Adir Cohen, Noam Mor et al.

Summary

Koshorek et al. reframe text segmentation, which splits a document into topically coherent contiguous parts, as supervised learning. Earlier systems relied on unsupervised heuristics, topic models or graph methods, mainly because labeled corpora were tiny or synthetic. The authors introduce WIKI-727K, a corpus of 727,746 English Wikipedia articles segmented automatically using each article's table of contents. Each sentence receives a binary label marking whether it closes a segment. Their model is hierarchical. A bidirectional LSTM with max-pooling builds a vector for each sentence, and a second bidirectional LSTM reads these vectors and predicts a boundary probability for each sentence. Boundaries are then chosen by thresholding. Evaluated with the Pk metric, the model clearly beats GraphSeg and a random baseline on Wikipedia text and comes close to human annotators on a 50-document sample. It is weaker on Choi's synthetic dataset and on the chemistry-themed ELEMENTS set. It also runs much faster than GraphSeg. The work shows that cheap, naturally structured data can support learned segmenters for real text, and both the corpus and the code are publicly released.

Key points

  • WIKI-727K corpus: 727,746 documents split 80/10/10 into train, development and test. It keeps the full hierarchical segmentation, sentences are split with NLTK's PUNKT tokenizer, and it is released at github.com/koomri/text-segmentation.
  • Training setup: the model learns only top-level segments and uses GoogleNews word2vec embeddings. The first segment of each article, typically a broad summary, is dropped during training.
  • Pk results on WIKI-50 (lower is better): the model scores 18.24, compared with GraphSeg at 63.56, the random baseline at 52.65 and humans at 14.97. On the full WIKI-727K test set the model reaches 22.13.
  • Mixed results on older benchmarks:
    • On CITIES, the model beats Chen et al. (2009), scoring 18.14 versus 22.1 with word-level Pk.
    • On ELEMENTS, it trails Chen et al., scoring 33.82 versus 20.1.
    • On CHOI, GraphSeg wins, scoring 5.6–7.2 versus 26.26.
  • Runtime: the model scales linearly with text length and averages 1.6 s per document on a laptop CPU, compared with 23.6 s for GraphSeg on WIKI-50.

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

🎓Tel Aviv UniversityTel Aviv · Israel67 works · 10,006 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 Omri Koshorek, Adir Cohen, Noam Mor, Michael Rotman and Jonathan Berant. 2018. Text Segmentation as a Supervised Learning Task. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n18-2075

@inproceedings{koshorek2018text,
  title = {{Text Segmentation as a Supervised Learning Task}},
  author = {Omri Koshorek, Adir Cohen, Noam Mor, Michael Rotman and Jonathan Berant},
  booktitle = {Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers)},
  year = {2018},
  doi = {10.18653/v1/n18-2075},
  url = {https://doi.org/10.18653/v1/n18-2075}
}

References

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

  1. Martin Riedl and Chris Biemann. 2012. Topictiling: a text segmentation algorithm based on lda. In Proceedings of ACL 2012 Student Research Workshop. Association for Computational Linguistics, pages 37–42. find
  2. Doug Beeferman, Adam Berger, and John Lafferty. 1999. Statistical models for text segmentation. Machine learning 34(1):177–210. find
  3. S. Bird, E. Loper, and E. Klein. 2009. Natural Language Processing with Python. OReilly Media Inc. find
  4. Masao Utiyama and Hitoshi Isahara. 2001. A statistical model for domain-independent text segmentation. In Proceedings of the 39th Annual Meeting on Association for Computational Linguistics. Association for Computational Linguistics, pages 499–506.
  5. D. Blei, A. Ng, and M. I. Jordan. 2003. Latent Dirichlet allocation. Journal of Machine Learning Research (JMLR) 3:993–1022.
  6. Thorsten Brants, Francine Chen, and Ioannis Tsochantaridis. 2002. Topic-based document segmentation with probabilistic latent semantic analysis. In Proceedings of the eleventh international conference on Information and knowledge management. ACM, pages 211–218. find
  7. Harr Chen, SRK Branavan, Regina Barzilay, and David R Karger. 2009. Global models of document structure using latent permutations. In Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics. Association for… find
  8. Freddy YY Choi. 2000. Advances in domain independent linear text segmentation. In Proceedings of the 1st North American chapter of the Association for Computational Linguistics conference. Association for Computational Linguistics, pages 26–33.
  9. Chris Fournier. 2013. Evaluating Text Segmentation using Boundary Edit Distance. In Proceedings of 51st Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, Stroudsburg, PA, USA, page to appear. find
  10. Goran Glavaš, Federico Nanni, and Simone Paolo Ponzetto. 2016. Unsupervised text segmentation using semantic relatedness graphs. Association for Computational Linguistics. find
  11. Marti A Hearst. 1994. Multi-paragraph segmentation of expository text. In Proceedings of the 32nd annual meeting on Association for Computational Linguistics. Association for Computational Linguistics, pages 9–16. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/koshorek-2018-text-segmentation-as-supervised-learning/