Equivalence between nonnegative tensor factorization and tensorial probabilistic latent semantic analysis

ABSTRACT

This paper establishes a connection between NMF and PLSA on multi-way data, called NTF and T-PLSA respectively. Two types of T-PLSA models are proven to be equivalent to non-negative PARAFAC and non-negative Tucker3. This paper also shows that by running NTF and T-PLSA alternatively, they can jump out of each other's local minima and achieve a better clustering solution.

REFERENCES

• 1: Yun Chi , Shenghuo Zhu , Yihong Gong , Yi Zhang, Probabilistic polyadic factorization and its application to personalized recommendation, Proceeding of the 17th ACM conference on Information and knowledge management, October 26-30, 2008, Napa Valley, California, USA [doi>10.1145/1458082.1458206]
• 2: Chris Ding , Tao Li , Wei Peng, On the equivalence between Non-negative Matrix Factorization and Probabilistic Latent Semantic Indexing, Computational Statistics & Data Analysis, v.52 n.8, p.3913-3927, April, 2008 [doi>10.1016/j.csda.2008.01.011]
• 3: Richard O. Duda , Peter E. Hart , David G. Stork, Pattern Classification (2nd Edition), Wiley-Interscience, 2000
• 4: Thomas Hofmann, Unsupervised learning by probabilistic latent semantic analysis, Machine Learning, v.42 n.1/2, p.177-196, Jan. 2001 [doi>10.1023/A:1007617005950]
• 5: D. D. Lee and S. H. Seung. Learning the parts of objects by non-negative matrix factorization. Nature, 401(6755):788--791, October 1999.
• 6: Wei Peng , Tao Li, Author-topic evolution analysis using three-way non-negative Paratucker, Proceedings of the 31st annual international ACM SIGIR conference on Research and development in information retrieval, July 20-24, 2008, Singapore, Singapore [doi>10.1145/1390334.1390521]
• 7: R.A.Harshman. Foundations of the parafac procedure: models and conditions for an 'explanatory' multi-modal factor analysis. UCLA working papers in phonetics 16, pages 1--84, 1970.
• 8: M. Shashanka, B. Raj, and P. Smaragdis. Probabilistic latent variable models as nonnegative factorizations. Computational intelligence and neuroscience, 2008.

INDEX TERMS

Primary Classification:
  I. Computing Methodologies
  I.5 PATTERN RECOGNITION
      I.5.3 Clustering
          Subjects: Algorithms

General Terms:
Algorithms, Experimentation, Performance

Keywords:
NTF, PLSA, multi-way data, tensor