Learning contextual rules for document understanding

We propose a supervised inductive learning approach for the problem of document understanding, that is, recognizing logical components of a document. For this purpose, FOCL and NDUBI/H, two systems that learn Horn clauses, have been employed. Several experimental results are reported and a critical view of the underlying independence assumption, made by almost all systems that learn from examples, is presented. This led us to redefine the problem of document understanding in terms of a new strategy of supervised inductive learning, called contextual learning. Experiments, in which a dependency hierarchy between concepts is defined, show that contextual rules increase predictive accuracy and decrease learning time for labelling problems, like document understanding. Encouraging results have been obtained when we tried to discover a linear dependency order by means of statistical methods