Gestion de l'incertitude et de l'imprécision dans un processus d'extraction de connaissances à partir des textes. (Uncertainty and imprecision management in a knowledge extraction process from unstructured texts)

Knowledge discovery and inference are concepts tackled in different ways in the scientific literature. Indeed, a large number of domains are interested such as : information retrieval, textual inference or knowledge base population. These concepts are arousing increasing interest in both academic and industrial fields, promoting development of new methods. This manuscript proposes an automated approach to infer and evaluate knowledge from extracted relations in unstructured texts. Its originality is based on a novel framework making it possible to exploit (i) the linguistic uncertainty thanks to an uncertainty detection method described in this manuscript (ii) a generated partial ordering of studied objects (e.g. noun phrases) taking into account of syntactic implications and a priori knowledge defined into taxonomies, and (iii) an evaluation step of extracted and inferred relations by selection models exploiting a specific partial ordering of relations. This partial ordering allows to compute some criteria in using information propagation rules in order to evaluate the belief associated to a relation in taking into account of the linguistic uncertainty. The proposed approach is illustrated and evaluated through the definition of a system performing question answering by analysing texts available on the Web. This case study shows the benefits of structuring processed information (e.g. using prior knowledge), the impact of selection models and the role of the linguistic uncertainty for inferring and discovering new knowledge. These contributions have been validated by several international and national publications and our pipeline can be downloaded at https ://github.com/PAJEAN/.