Mouna Kamel

A semi-automatic approach for building ontologies from acollection of structured web documents

Many collections of structured documents are available on the web. The collection generally describes the characteristics of entities from a single type, where each page describes one entity. These documents are adequate knowledge sources for building ontologies. As they benefit from a strong and shared layout, they contain less well written text than plain text files but their architecture is very meaningful. Classical linguistic-based methods for identifying concepts and relations are no longer appropriate for analyzing them.The approach we propose in this paper exploits various properties of such documents, combining layout/formatting analysis and linguistic analysis, and using semantic annotation.

Discovering Hypernymy Relations using Text Layout

Hypernymy relation acquisition has been widely investigated, especially because taxonomies, which often constitute the backbone structure of semantic resources are structured using this type of relations. Although lots of approaches have been dedicated to this task, most of them analyze only the written text. However relations between not necessarily contiguous textual units can be expressed, thanks to typographical or dispositional markers. Such relations, which are out of reach of standard NLP tools, have been investigated in well specified layout contexts. Our aim is to improve the relation extraction task considering both the plain text and the layout. We are proposing here a method which combines layout, discourse and terminological analyses, and performs a structured prediction. We focused on textual structures which correspond to a well defined discourse structure and which often bear hypernymy relations. This type of structure encompasses titles and sub-titles, or enumerative structures. The results achieve a precision of about 60%.

Eliciting hierarchical structures from enumerative structures for ontology learning

Some discourse structures such as enumerative structures have typographical, punctuational and laying out characteristics which (1) make them easily identifiable and (2) convey hierarchical relations which provide ontology fragments clues. This study will try to show how these textual objects can be exploited in order to considerably improve the process of ontology enrichment from text.

Taking Advantage of Discursive Properties for Validating Hierarchical Semantic Relations from Parallel Enumerative Structures

This paper presents an approach for automatically validating candidate hierarchical relations extracted from parallel enumerative structures. It relies on the discursive properties of these structures and on the combination of resources of different nature, a semantic network and a distributional resource. The results show an accuracy of between 0.50 and 0.67, with a gain of 0.11 when combining the two resources.

A supervised machine learning approach for taxonomic relation recognition through non-linear enumerative structures

Improving relation extraction process requires to have a better insight of the proper text or to use external resources. Our work lies in the first term of this alternative, and aim at extending works about semantic relation identification in texts for building taxonomies which constitute the backbone of ontologies on which Semantic Web applications are built. We consider a specific discursive structure, the enumerative structure, as it bears explicit hierarchical knowledge. This structure is expressed with the help of lexical or typo-dispositional markers whose role is to introduce hierarchical levels between its components. Typo-dispositional markers are unfortunately not integrated into most parsing systems used for information extraction tasks. In order to extend the taxonomic relation identification process, we thus propose a method for recognizing this relation through enumerative structures which benefit from typo-dispositional markers (we called them non-linear enumerative structures). Our method is based on supervised machine learning. Two strategies have been applied: a linear classification with a MaxEnt and a non-linear one with a SVM. The results obtained in each of these approaches are close, with respectively an F1 of 81.25% and of 81.77%.