Learning (k, l)-Contextual Tree Languages for Information Extraction

Learning regular languages from positive examples only is known to be infeasible. A common solution is to define a learnable subclass of the regular languages. In the past, this has been done for regular string languages. Using ideas from those techniques, we define a learnable subclass of regular unranked tree languages, called the (k,l)-contextual tree languages. We describe the use of this subclass to induce wrappers for Information Extraction from structured documents, such as web pages. Experiments show that our algorithm is able to learn from very few data, and compares favorably to similar state of the art approaches. Learning (k,l)-Contextual Tree Languages for Information Extraction Stefan Raeymaekers, Maurice Bruynooghe, and Jan Van den Bussche 1 K.U.Leuven, Dept. of Computer Science, Celestijnenlaan 200A, B-3001 Leuven, {stefanr,maurice}@cs.kuleuven.ac.be 2 University of Limburg, Dept. WNI, Universitaire Campus, B-3590 Diepenbeek, [email protected] Abstract. Learning regular languages from positive examples only is known to be infeasible. A common solution is to define a learnable subclass of the regular languages. In the past, this has been done for regular string languages. Using ideas from those techniques, we define a learnable subclass of regular unranked tree languages, called the (k,l)-contextual tree languages. We describe the use of this subclass to induce wrappers for Information Extraction from structured documents, such as web pages. Experiments show that our algorithm is able to learn from very few data, and compares favorably to similar state of the art approaches. Learning regular languages from positive examples only is known to be infeasible. A common solution is to define a learnable subclass of the regular languages. In the past, this has been done for regular string languages. Using ideas from those techniques, we define a learnable subclass of regular unranked tree languages, called the (k,l)-contextual tree languages. We describe the use of this subclass to induce wrappers for Information Extraction from structured documents, such as web pages. Experiments show that our algorithm is able to learn from very few data, and compares favorably to similar state of the art approaches.