La réutilisation des connaissances en supervision de réseaux de télécommunications
Institution:
Paris 8Disciplines:
Directors:
Abstract EN:
This doctoral thesis is based on a combined psychological and ergonomic approach. It concerns research on analogical reasoning and on episodic memory retrieval. More precisely this research focused on knowledge reuse in telephone traffic monitoring in order to better design support tools. Case-based reasoning (cbr) in artificial intelligence has been studied in this thesis to see if it could possibly be used as a support system. Task analyses, activity analyses, behavior trace analyses, incident traces and an experiment using an incident categorisation task were conducted. These methodological approaches showed that two main factors were involved in traffic monitoring from the diagnosis of the incidents to their categorisation. These are the multiplicity of the task goals, and the heterogeneity of the operators' episodic knowledge. As postulated by the " episodic memory " theory on specific encoding, the encoding context determines knowledge retrieval and more specifically in work situations where the context is linked to the task and its goals but also to the work team and to the knowledge shared by the operators. The similarity between incidents has been shown to be " situated " in the context of a goal oriented activity and for each specific operator. It then appeared impossible to use a cbr support tool that would not take into account the operators' variability. The special relations linking the task and the operator that intervene in the process of describing and retrieving the incidents, lead to the study of a formal description. In fact these formal description should be significant, functional to the operators.
Abstract FR:
Cette these a pour theme la reutilisation des connaissances en supervision de reseaux de telecommunications. Elle s'incrit dans le cadre general des etudes en psychologie sur la recuperation en memoire "episodique" et sur l'analogie, et en ergonomie sur la conception de systemes d'aide. L'approche "raisonnement a partir de cas" (rc) en intelligence artificielle est etudiee du point de vue de son apport en tant qu'aide. Une analyse de l'activite et des traces (incidents traites) ainsi qu'une experience de regroupements d'incidents ont permis de mettre a jour deux facteurs agissant depuis le diagnostic jusqu'a la categorisation des incidents lors de leur consignation : la multiplicite des finalites liees aux taches et l'heterogeneite des connaissances episodiques possedees par les operateurs. Comme postule dans la theorie de la memoire episodique sur l'encodage specifique, le contexte d'encodage conditionne la recuperation et notamment dans les situations de travail, ou le contexte est certes lie a la tache et a ses finalites, mais aussi au collectif de travail et a son referentiel commun. La similarite entre deux incidents etant "situee" dans le cadre d'une activite finalisee et pour un operateur particulier, il est impossible d'utiliser un systeme d'aide de type rc qui ne tiendrait pas compte de la variabilite des operateurs. La double dependance operateur/tache intervenant dans les processus de description et de recuperation des incidents conduit a poser le probleme d'une description formelle