PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Stacked dependency networks for layout document structuring
Loic Lecerf and Boris Chidlovskii
In: ACM SAC 2008, 16-20 March 2008, Fortaleza, Brazil.

Abstract

We address the problems of structuring and annotation of layout-oriented documents. We model the annotation problems as the collective classification on graph-like structures with typed instances and links that captu re the domain-specific knowledge. We use the relational dependency networks (RDNs) for the collective infere nce on the multi-typed graphs. %to model the collective classification of typed elements and links between them, on the base of their chara cteristics and inter-dependencies. We then describe a variant of RDNs where a stacked approximation replaces the Gibbs sampling in order to acc elerate the inference. %For the structure learning in the stacked RDNs, long distance relationships between elements are detected u sing the greedy algorithm {\it on each level of the stack}; the algorithm guarantees the structural consiste ncy of the resulting stacked RDNs. We report results of evaluation tests for both the Gibbs sampling and stacking inference on two document str ucturing examples.

EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
ID Code:5328
Deposited By:Boris Chidlovskii
Deposited On:24 March 2009