Search

Ontology alignment plays a key role in achieving interoperability on the semantic Web. Inspired by the success of word embedding techniques in several NLP tasks, we propose a new ontology alignment approach based on the combination of word embedding and the radius measure. We tested our system on the OAEI (http://oaei.ontologymatching.org/) conference track and then applied it to aligning ontologies in a real-world case study. The experimental results show that using word embedding and the radius measure make it possible to determine, with good accuracy, not only equivalence relations, but also hierarchical relations between concepts.

In this paper we first present the international multidisciplinary research network Zoomathia, which aims at studying the transmission of zoological knowledge from Antiquity to Middle Ages through varied resources, and considers especially textual information, including compilation literature such as encyclopaedias. We then present a preliminary work in the context of Zoomathia consisting in i extracting pertinent knowledge from mediaeval texts using Natural Language Processing NLP methods, ii semantically enriching semi-structured zoological data and publishing it as an RDF dataset and its vocabulary, linked to other relevant Linked Data sources, and iii reasoning on this linked RDF data to help epistemologists, historians and philologists in their analysis of these ancient texts. This paper is an extended and updated version ofi¾ź[13].

This CIFRE doctoral thesis is part of a collaborative research project between the I3S laboratory of the University of Côte d'Azur and the Silex company, and addresses the field of recommendation systems. Silex is a start-up that develops a Software-as-a-Service sourcing tool that allows companies to provide a description of their professional activities, their offers and/or the services they are looking for in natural language (currently French).In this context, the objective of this thesis is to propose a decision support system by exploiting the semantic knowledge that are extracted from the textual descriptions of requests for services and providers, in order to recommend relevant providers for a service request.The contributions of this thesis are the following. First, we proposed a vocabulary for the sourcing field by reusing and integrating existing vocabularies, in order to semantically annotate the textual descriptions of providers and requests for services. Second, we proposed an automatic alignment method to establish the correspondence between different concepts of the considered vocabularies. This approach is based on rules exploiting embedding space and measurements on groups of labels to discover the relationships between concepts. Third, we proposed an algorithm for extracting named entities from the textual descriptions of service requests and providers, and an algorithm for semantic annotation of these descriptions, based on the linking of the extracted entities with the concepts of the defined vocabulary.Fourth, we proposed a provider recommendation algorithm that exploits these knowledges extracted.Finally, we studied the contribution of using ontological knowledge to improve our decision support system for the sourcing domain in order to recommend relevant providers for a service request.The contributions of this thesis are the following. First, we proposed a vocabulary for the sourcing field in order to semantically annotate the textual descriptions of providers and requests for services. This vocabulary was built by reusing and integrating existing vocabularies. Second, we proposed an automatic alignment method to establish the correspondence between different concepts of the considered vocabularies. This approach is based on rules exploiting embedding space and measurements on groups of labels to discover the relationships between concepts. Third, we proposed an algorithm for extracting named entities from the textual descriptions of service requests and providers, and an algorithm for semantic annotation of these descriptions, based on the linking of the extracted entities with the concepts of the defined vocabulary.Fourth, we proposed a provider recommendation algorithm that exploits these knowledge extracted.Finally, we studied the contribution of using ontological knowledge to improve our decision support system for the sourcing domain.; Cette thèse de doctorat CIFRE s'inscrit dans le cadre d'un projet de recherche collaboratif entre le laboratoire I3S de l'Université Côte d'Azur et la société Silex et aborde le domaine des systèmes de recommandation. Silex est une start-up qui développe un outil de sourcing Software-as-a-Service permettant aux entreprises de fournir une description de leurs activités professionnelles, de leurs offres et/ou des services qu'elles recherchent en langue naturelle (actuellement le français).Dans ce contexte, l'objectif de cette thèse est de proposer un système d'aide à la décision en exploitant les connaissances sémantiques extraites à partir des descriptions textuelles des demandes de prestation et des prestataires, afin de recommander des prestataires pertinents pour une demande de prestation.Les contributions de cette thèse sont les suivantes. Premièrement, nous avons proposé un vocabulaire pour le domaine du sourcing en réutilisant et en intégrant des vocabulaires existants, afin d'annoter sémantiquement les descriptions textuelles des prestataires et des demandes de prestation. Deuxièmement, nous avons proposé une méthode d’alignement automatique afin d'établir la correspondance entre différents concepts des vocabulaires considérés. Cette approche se base sur des règles exploitant l'espace des plongements lexicaux et des mesures sur des groupes d'étiquettes pour découvrir les relations entre concepts. Troisièmement, nous avons proposé un algorithme d'extraction des entités nommées à partir des descriptions textuelles des demandes de prestation et des prestataires et un algorithme d'annotation sémantique de ces descriptions, basé sur le liage des entités extraites avec les concepts du vocabulaire défini.Quatrièmement, nous avons proposé un algorithme de recommandation de prestataires qui exploite ces annotations sémantiques.Finalement, nous avons étudié l'apport de l'utilisation de connaissances ontologiques afin d'améliorer notre système d'aide à décision pour le domaine du sourcing.

International audience; In this paper we first present the international multidisciplinary research network Zoomathia, which aims the study of the transmission of zoological knowledge from Antiquity to Middle Ages through varied resources, and considers especially textual information, including compilation literature such as encyclopaedias. We then present a preliminary work in the context of Zoomathia consisting in (i) extracting pertinent knowledge from mediaeval texts using Natural Language Processing (NLP) methods, (ii) semantically enriching semi-structured zoological data and publishing it as an RDF dataset and its vocabulary, linked to other relevant Linked Data sources, and (iii) reasoning on this linked RDF data to help epistemologists, historians and philologists in their analysis of these ancient texts.

National audience; In a number of areas, companies are often faced with the task of dealing with large amounts of textual customers' requests. Automating information extraction like key phrases from customers' requests can help to accelerate the processing process. Silex France is currently facing this challenge in the context of processing sourcing requests.In this article, we share our sequence labeling results based on a hybrid method Bi-LSTM-CRF, in an industrial context. This work was integrated in the B2B Silex platform for service providers recommendation. Experiments with the B2B Silex platform data show that, with a good choice of features to extract and optimal choice of hyper-parameters, the combination of the Bi-LSTM and CRF helps to achieve good results even in a context of small data. Indeed, the textual content processed is in the form of complete sentences generated by users, and thus is subject to typing errors. To handle this type of data we combine several types of extracted features describing the textual content such as: (i) semantics, (ii) syntax, (iii) word characters, (iv) position of words.; Dans un certain nombre de domaines, les entreprises sont souvent confrontées à la tâche de traiter au quotidien des quantités importantes de demandes textuelles. L'extraction automatique des informations clés à partir des demandes clients, peut aider à accélérer le processus de traitement. Silex France est aujourd'hui confrontée à ces enjeux dans le cadre du traitement des demandes de sourcings. Dans cet article, nous partageons nos résultats d'étique-tage de séquences en nous basant sur une méthode hybride BiLSTM-CRF, dans un contexte industriel. Le travail est in-tégré dans la plate-forme B2B Silex pour la recommanda-tion des prestataires de services. Les expériences faites sur les données de la plateforme B2B Silex montrent qu'avec un bon choix de features à extraire et des hyperparamètres, la combinaison du modèle Bi-LSTM-CRF permet de réus-sir l'extraction d'infomation à partir des demandes tex-tuelles, même dans un contexte de petites données (small data). En effet, le contenu textuel traité est sous forme de phrases complètes générées par des utilisateurs, et est ainsi exposé à des erreurs de frappe. Pour gérer ce type de don-nées, nous combinons plusieurs types de features extraites décrivant le contenu textuel tels que : (i) la sémantique, (ii) la syntaxe, (iii) les caractères des mots, (iv) la position des mots.