Towards a hybrid approach for supervising interactive adaptive systems - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

Towards a hybrid approach for supervising interactive adaptive systems

Résumé

The analysis of users' behavior (for example the visitors of a website or tourists using a dedicated mobile application) tackle a common problem: the rapid extraction of path's patterns. These patterns will go back to a user profile characterization and their behaviors. The goal is to achieve a dynamic content adaptation according to these behaviors and the specific context of navigation. Various works were interested in how to acquire and dynamically analyze navigation datas through contents by extracting relevant information about users's interest or motivation and proposing dynamic adaptations rules relevant for the specific situation. Two approaches allow to address this problem: one based on machine learning techniques (i.e. reinforcement algorithms) for extracting representative information on the basis of past navigations logs, the other is based on the dynamic observation of the behavior's user and a set of scenario of modification rules. Therefore our work intends to develop a generic framework, based on a formal model, for analyzing users' behavior and dynamic content adaptation using a hybrid approach combining learning and real-time observation/scenario modification. 1.1 State of the art There are already a huge number of approaches related to the dynamic adaptation for interactive sto-rytelling. We can classify them into three classes. First, scripted approaches [Vega and L., 2003], [Rempulski et al., 2009] are based on an expert system controlling the user experience. The scripted approaches generate the narrative that is closest to the expectations of the latter based on the observation that is made and compared to a set of predefined rules. Modeling such an architecture is a complex task as all the possible paths for the user must be defined by the designer. Oriented agent approaches aim to the emergence of narratives based on a set of agents' behaviors [Mateas and Stern, 2003]. The modular-ity of this way of modeling the system makes it easier to design but this approach makes it more difficult to control the quality and consistency of the generated model, what scripted approaches guarantee by their very structure. Finally, hybrid approaches try to find the balance between the user's freedom, and thus a high level of adaptation obtained with oriented agent approach, and the respect of the designer's frame (designed with scripted approaches). [Brian Magerko, 2003] offers such an hybrid interactive sto-rytelling system, where the author specifies a generic framework without limiting the user's actions. The quality of the execution is carried out by a manager that dynamically analyzes the behavior of users to detect unanticipated behaviors. The major difficulty stands in the representation of all the possible interactions between the user and the system. The model must garantee a high level of modularity, reusability and want be scalable enough to meet potential changes. In addition, supervision models requires having a formal model from which
Fichier principal
Vignette du fichier
movep.pdf (186.9 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01670557 , version 1 (21-12-2017)

Identifiants

  • HAL Id : hal-01670557 , version 1

Citer

Damien Mondou, Armelle Prigent, Arnaud Revel, Nicolas Rempulski. Towards a hybrid approach for supervising interactive adaptive systems. MOVEP, Jun 2016, Gênes, Italy. ⟨hal-01670557⟩
327 Consultations
81 Téléchargements

Partager

Gmail Facebook X LinkedIn More