Section: Software and Platforms

Knowledge-Based Systems and Semantic Web Systems

The Kasimir System for Decision Knowledge Management

Participants : Nicolas Jay, Jean Lieber [contact person] , Amedeo Napoli, Thomas Meilender.

classification-based reasoning, case-based reasoning, edition and maintenance of knowledge, decision knowledge management, semantic portal

The objective of the Kasimir system is decision support and knowledge management for the treatment of cancer. A number of modules have been developed within the Kasimir system for editing of treatment protocols, visualization, and maintenance. Kasimir is developed within a semantic portal, based on OWL. KatexOWL (Kasimir Toolkit for Exploiting OWL Ontologies, http://katexowl.loria.fr ) has been developed in a generic way and is applied to Kasimir. In particular, the user interface EdHibou of KatexOWL is used for querying the protocols represented within the Kasimir system.

The software CabamakA (case base mining for adaptation knowledge acquisition) is a module of the Kasimir system. This system performs case base mining for adaptation knowledge acquisition and provides information units to be used for building adaptation rules. Actually, the mining process in CabamakA is implemented thanks to a frequent close itemset extraction module of the Coron platform (see § 5.1.1 ).

The Oncologik system is a collaborative editing tool aiming at facilitating the management of medical guidelines [49] , [48] . Based on a semantic wiki, it allows the acquisition of formalized decision knowledge. A production version was released this year (http://www.oncologik.fr/ ). Oncologik also includes a graphical decision tree editor, KcatoS [61] .

Taaable: a system for retrieving and creating new cooking recipes by adaptation

Participants : Valmi Dufour-Lussier, Emmanuelle Gaillard, Laura Infante Blanco, Florence Le Ber, Jean Lieber, Amedeo Napoli, Emmanuel Nauer [contact person] .

knowledge acquisition, ontology engineering, semantic annotation, case-based reasoning, hierarchical classification, text mining

Taaable is a system whose objectives are to retrieve textual cooking recipes and to adapt these retrieved recipes whenever needed. Suppose that someone is looking for a “leek pie” but has only an “onion pie” recipe: how can the onion pie recipe be adapted?

The Taaable system combines principles, methods, and technologies of knowledge engineering, namely case-based reasoning (CBR), ontology engineering, text mining, text annotation, knowledge representation, and hierarchical classification. Ontologies for representing knowledge about the cooking domain, and a terminological base for binding texts and ontology concepts, have been built from textual web resources. These resources are used by an annotation process for building a formal representation of textual recipes. A CBR engine considers each recipe as a case, and uses domain knowledge for reasoning, especially for adapting an existing recipe w.r.t. constraints provided by the user, holding on ingredients and dish types.

The Taaable system is available since 2008 on line at http://taaable.fr , but is constantly evolving. This year, Taaable has been extended by two new features, both concerning knowledge acquisition.

The first feature uses closed itemsets for extracting adaptation knowledge in order to better adapt recipes. A first approach integrates a previous work about adaptation rule extraction  [93] into a collaborative environment, in which humans and machines may now collaborate to better acquire adaptation rules [38] . This environment integrates also the results of a new work on knowledge extraction where specific cooking adaptation rules that can be applied to a single recipe, are generalized using close itemsets into generic adaptation rules, to make them usable on other recipes [60] .

The second feature addresses the improvement of the formal representation of the preparation part of recipes, using a semi-automatic annotation process [59] . In Taaable, the procedural text describing the preparation is formalized in a graph, where cooking actions and ingredients, among others, are represented as vertexes, and semantic relations between those, shown as arcs. As the automatic annotation process that transforms, using natural language processing, a procedural text into a graph, produces incomplete annotation (disconnected graphs) or other annotation errors, a validating and correcting step is required. A specific graphical interface has been built to provide the users with a way to correct the graph representation of the cooking process, improving at the same time the quality of the knowledge about cooking procedures.

Tuuurbine: a generic ontology guided case-based inference engine

Participants : Laura Infante Blanco, Jean Lieber, Emmanuel Nauer [contact person] .

case-based reasoning, inference engine, knowledge representation, ontology engineering, semantic web

The experience acquired since 5 years with the Taaable system conducted to the creation of a generic cased-based reasoning system, whose reasoning procedure is based on a domain ontology. This new system, called Tuuurbine, takes into account the retrieval step, the case base organization, but also an adaptation procedure which is not addressed by other generic case-based reasoning tools. Moreover, Tuuurbine is built over semantic web standards that will ensure facilities for being plugged over data available on the web. The domain knowledge is considered to be represented in a RDF store, which could be additionally be interfaced with a semantic wiki, in order to benefit from the collaborative edition and management of the knowledge involved in the reasoning system (cases, ontology, adaptation rules). This development is support by an Inria ADT funding.