Advertisement

Theoretical Analysis and Evaluation of Topic Graph Based Transfer Learning

  • Tetsuya Yoshida
  • Hiroki Ogino
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 8210)

Abstract

Various research efforts have been invested in machine learning and data mining for finding out patterns from data. However, even when some knowledge may be learned in one domain, it is often difficult to re-use it for another domain with different characteristics. Toward effective knowledge transfer between domains, we proposed a transfer learning method based on our transfer hypothesis that two domains have similar feature spaces. A graph structure called a topic graph is constructed by using the learned features in one domain, and the graph is used as a regularization term. In this paper we present a theoretical analysis of our approach and prove the convergence of the learning algorithm. Furthermore, the performance evaluation of the method is reported over document clustering problems. Extensive experiments are conducted to compare with other transfer learning algorithms. The results are encouraging, and show that our method can improve the performance by transferring the learned knowledge effectively.

Keywords

Regularization Term Target Domain Normalize Mutual Information Transfer Learning Source Domain 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Cai, D., He, X., Wu, X., Han, J.: Non-negative matrix factorization on manifold. In: Proc. ICDM 2008, pp. 63–72 (2008)Google Scholar
  2. 2.
    Cover, T., Thomas, J.: Elements of Information Theory. Wiley (2006)Google Scholar
  3. 3.
    Dai, W., Xue, G.-R., Yang, Q., Yu, Y.: Co-clustering based classification for out-of-domain documents. In: Proc. KDD 2007, pp. 210–219 (2007)Google Scholar
  4. 4.
    Dhillon, J., Modha, D.: Concept decompositions for lage sparse text data using clustering. Machine Learning 42, 143–175 (2001)CrossRefzbMATHGoogle Scholar
  5. 5.
    Ding, C., Li, T., Peng, W., Park, H.: Orthogonal nonnegative matrix tri-factorizations for clustering. In: Proc. KDD 2006, pp. 126–135 (2006)Google Scholar
  6. 6.
    Gao, J., Fan, W., Jiang, J., Han, J.: Knowledge transfer via multiple model local structure mapping. In: Proc. KDD 2008, pp. 283–291 (2008)Google Scholar
  7. 7.
    Lee, D.D., Seung, H.S.: Algorithms for non-negative matrix factorization. In: Proc. NIPS 2001, pp. 556–562 (2001)Google Scholar
  8. 8.
    Ling, X., Dai, W., Xue, G., Yang, Q., Yu, Y.: Spectral domain-transfer learning. In: Proc. KDD 2008, pp. 488–496 (2008)Google Scholar
  9. 9.
    Ogino, H., Yoshida, T.: Topic graph based non-negative matrix factorization for transfer learning. In: Kryszkiewicz, M., Rybinski, H., Skowron, A., Raś, Z.W. (eds.) ISMIS 2011. LNCS, vol. 6804, pp. 260–269. Springer, Heidelberg (2011)CrossRefGoogle Scholar
  10. 10.
    Pan, S.J., Yang, Q.: A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering 22(10), 1345–1359 (2010)CrossRefGoogle Scholar
  11. 11.
    Strehl, A., Ghosh, J.: Cluster ensembles — a knowledge reuse framework for combining multiple partitions. J. Machine Learning Research 3(3), 583–617 (2002)MathSciNetGoogle Scholar
  12. 12.
    von Luxburg, U.: A tutorial on spectral clustering. Statistics and Computing 17(4), 395–416 (2007)MathSciNetCrossRefGoogle Scholar
  13. 13.
    Xu, W., Liu, X., Gong, Y.: Document clustering based on non-negative matrix factorization. In: Proc. SIGIR 2003, pp. 267–273 (2003)Google Scholar
  14. 14.
    Zhuang, F., Luo, P., Xiaong, H., He, Q., Xiong, Y., Shi, Z.: Exploiting associations between word clusters and document classes for cross-domain text categorization. In: Proc. SDM 2010, pp. 13–24 (2010)Google Scholar

Copyright information

© Springer International Publishing Switzerland 2013

Authors and Affiliations

  • Tetsuya Yoshida
    • 1
  • Hiroki Ogino
    • 1
  1. 1.Graduate School of Information Science and TechnologyHokkaido UniversitySapporoJapan

Personalised recommendations