Separating Precision and Mean in Dirichlet-Enhanced High-Order Markov Models

  • Rikiya Takahashi
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 4701)

Abstract

Robustly estimating the state-transition probabilities of high-order Markov processes is an essential task in many applications such as natural language modeling or protein sequence modeling. We propose a novel estimation algorithm called Hierarchical Separated Dirichlet Smoothing (HSDS), where Dirichlet distributions are hierarchically assumed to be the prior distributions of the state-transition probabilities. The key idea in HSDS is to separate the parameters of a Dirichlet distribution into the precision and mean, so that the precision depends on the context while the mean is given by the lower-order distribution. HSDS is designed to outperform Kneser-Ney smoothing especially when the number of states is small, where Kneser-Ney smoothing is currently known as the state-of-the-art technique for N-gram natural language models. Our experiments in protein sequence modeling showed the superiority of HSDS both in perplexity evaluation and classification tasks.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Chen, S., Goodman, J.: An empirical study of smoothing techniques for language modeling. Technical Report TR-10-98, Harvard Computer Science (1998)Google Scholar
  2. 2.
    Ganapathiraju, M., Manoharan, V., Klein-Seetharaman, J.: BLMT: Statistical sequence analysis using n-grams. Applied Bioinformatics 3 (November 2004)Google Scholar
  3. 3.
    Netzer, O., Lattin, J.M., Srinivasan, V.: A Hidden Markov Model of Customer Relationship Dynamics. Stanford GSB Research Paper (July 2005)Google Scholar
  4. 4.
    Kneser, R., Ney, H.: Improved backing-off for m-gram language modeling. In: Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, vol. 1, pp. 181–184 (May 1995)Google Scholar
  5. 5.
    Pitman, J., Yor, M.: The two-parameter Poisson-Dirichlet distribution derived from a stable subordinator. The Annals of Probability 25(2), 855–900 (1997)MATHCrossRefMathSciNetGoogle Scholar
  6. 6.
    Goldwater, S., Griffiths, T., Johnson, M.: Interpolating between types and tokens by estimating power-law generators. In: Advances in Neural Information Processing Systems (NIPS), vol. 18 (2006)Google Scholar
  7. 7.
    Teh, Y.W.: A Bayesian interpretation of interpolated Kneser-Ney. Technical Report TRA2/06, School of Computing, National University of Singapore (2006)Google Scholar
  8. 8.
    Teh, Y.W.: A hierarchical Bayesian language model based on Pitman-Yor processes. In: Proceedings of the Annual Meeting of the Association for Computational Linguistics, vol. 44 (2006)Google Scholar
  9. 9.
    MacKay, D.J.C., Peto, L.: A hierarchical Dirichlet language model. Natural Language Engineering 1(3), 1–19 (1994)Google Scholar
  10. 10.
    Minka, T.: Estimating a Dirichlet distribution. Technical report, Microsoft Research (2003)Google Scholar
  11. 11.
    Minka, T.: Beyond Newton’s method. Technical report, Microsoft Research (2000)Google Scholar
  12. 12.
    Ney, H., Essen, U., Kneser, R.: On structuring probabilistic dependences in stochastic language modeling. Computer, Speech, and Language 8, 1–38 (1994)CrossRefGoogle Scholar
  13. 13.
    Witten, I.H., Bell, T.C.: The zero-frequency problem: Estimating the probabilities of novel events in adaptive text compression. IEEE Transactions on Information Theory 37(4), 1085–1094 (1991)CrossRefGoogle Scholar
  14. 14.
    Lewis, D.D.: Reuters-21578 text categorization test collection distribution 1.0 (1997), Available at http://www.daviddlewis.com/resources/testcollections/reuters21578/
  15. 15.
    Guo, T., Sun, Z.: Dbsubloc: Database of protein subcellular localization (2005), Available at http://www.bioinfo.tsinghua.edu.cn/~guotao/

Copyright information

© Springer-Verlag Berlin Heidelberg 2007

Authors and Affiliations

  • Rikiya Takahashi
    • 1
  1. 1.IBM Tokyo Research Laboratory, 1623-14 Shimo-tsuruma, Yamato-shi, Kanagawa 242-8502Japan

Personalised recommendations