Reference:

Juha Raitio, Tapani Raiko, and Timo Honkela. Hybrid bilinear and trilinear models for exploratory analysis of three-way Poisson counts. In Proceedings of ICANN 2012, 22nd International Conference on Artificial Neural Networks, Volume II, pages 475–482, 2012.

Abstract:

We propose a probabilistic model class for the analysis of three-way count data, motivated by studying the subjectivity of language. Our models are applicable for instance to a data tensor of how many times each subject used each term in each context, thus revealing individual variation in natural language use. As our main goal is exploratory analysis, we propose hybrid bilinear and trilinear models with zero-mean constraints, separating modeling the simpler and more complex phenomena. While helping exploratory analysis, this approach leads into a more involved model selection problem. Our solution by forward selection guided by cross-validation likelihood is shown to work reliably on experiments with synthetic data.

Suggested BibTeX entry:

@inproceedings{RaitioRaikoHonkela12,
    author = {Juha Raitio and Tapani Raiko and Timo Honkela},
    booktitle = {Proceedings of ICANN 2012, 22nd International Conference on Artificial Neural Networks, Volume II},
    language = {eng},
    pages = {475-482},
    title = {Hybrid Bilinear and Trilinear Models for Exploratory Analysis of Three-Way {P}oisson Counts},
    year = {2012},
}