Publication: Sistema multi-agente basado en contexto, localización y reputación para dominios de inteligencia artificial
Loading...
Identifiers
Publication date
2012
Defense date
2012-10-31
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Las investigaciones en Inteligencia Ambiental (denominada también Computación Ubicua) utilizando tecnologías inalámbricas han crecido en los últimos años a pasos agigantados. El término Inteligencia Ambiental es un nuevo concepto adoptado para hacer referencia a entornos inteligentes, en donde la computadora y los dispositivos móviles asisten al usuario en sus actividades cotidianas. Son entornos con un gran despliegue de diferentes tecnologías, invisibles al usuario, que hacen posible la provisión de servicios personalizados. Los agentes inteligentes son un paradigma especial dentro de la Inteligencia Ambiental. La principal contribución de este trabajo consiste en el diseño metodológico de un sistema basado en contexto (incluyendo servicios de localización) utilizando agentes inteligentes, que da soporte a las necesidades de información de estas nuevas tecnologías, para dominios heterogéneos. Los agentes facilitan la comunicación y las interacciones entre los usuarios del sistema. Para hacer factible este intercambio de mensajes, es necesario contar con una ontología, la cual posibilita, no tan solo la comunicación entre los agentes intervinientes, sino que además, permite a los agentes razonar sobre el contexto en el que se sitúan. En este sentido, el uso de información geográfica permite inferir información de alto nivel sobre dónde se encuentra un agente. Las preferencias de los usuarios permiten seleccionar los mejores servicios en cada escenario de Inteligencia Ambiental. Esta personalización es particularmente adecuada en los sistemas multiagentes con capacidades de aprendizaje. En base a esta concepción, se aborda la adquisición de perfiles de usuarios, de forma no intrusiva, a través de técnicas de computación evolutiva. La adquisición de las preferencias del usuario hacen posible que se le presente al usuario una lista de actividades priorizada por realizar, según el dominio en el que se encuentre a efectos de maximizar factores que le sean relevantes. Además, en base a estos perfiles, los agentes proveedores realizan sus ofertas. Finalmente, las interacciones usuarios-proveedores en un dominio dinámico, requieren de cierto grado de confianza, el cual se adquiere a través del uso de sistemas de confianza y reputación. En este sentido, se presenta CALoR, un modelo de reputación basado en contexto y localización, que permite a los agentes usuarios del sistema realizar recomendaciones a otros agentes usuarios, considerando sus experiencias pasadas, y también aspectos espacio-temporales. Se asume que la recomendación enviada por un agente usuario sobre un agente proveedor es más fiable cuando han interactuado recientemente, y el agente usuario ha podido evaluar el servicio provisto a una distancia considerablemente corta. Resumiendo, en las páginas siguientes, se presentará al lector, una arquitectura multi-agente basada en contexto y localización para dominios heterogéneos de inteligencia ambiental. Se explicará en detalle la construcción del perfil de usuario utilizando algoritmos genéticos para la provisión de servicios personalizados, y se presentará el modelo de reputación CALoR para afianzar las relaciones entre los agentes de la arquitectura. Por cada módulo, se presentan los experimentos realizados y los resultados obtenidos, que demuestran la robustez de la propuesta global. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Researches on Ambient Intelligent (also known as Ubiquitous Computing) using wireless technologies have increased over the last few years. Ambient Intelligence is a new concept adopted for referencing an intelligent environment, where the user is assist by the computer and his mobile device wherever he is. Several technologies are necessaries in these environments to provide personalized services. All immersed devices are invisibles to the user. Ambient Intelligence is an ideal workspace for agents because of autonomous distributed and proactive agent nature. The main contribution of this Thesis consists of designing a methodological approach a Context Aware System (involving location services) using Agents that can be used in very different domains. We review several scenarios to define a multi-agent architecture, that support the information needs of these new technologies, that will be used in heterogeneous domain. We use ontology to facilitate the communication between agents. Furthermore, this ontology allows agent’s reasoning. In this manner, the use of geographic information allows to infer high level information about the real place where the agent is. In Ambient Intelligence scenarios, user preferences allow to select the best services for each user. This personalization is particularly suitable in a multi-agent system with learning capabilities. Based on this, and on the improvement of the mobile technologies, we developed an algorithm that allows learning the user profile. A list of prioritized services would therefore be presented to the user that takes into account user interests. Also, the providers agents could make their offers. Finally, users-providers interactions that take place in a dynamical domain needs a certain trust. Trust is build using trust and reputation systems. In this manner, we present CALoR reputation model that is a reputation model based-on context and location. This model allows the user agents of the systems to make recommendations from each other. CALoR take into account the past interactions of the agents and their spatial-temporal issues. We assume that a recommendation send by a user agent over a provider agent is more reliable when the agents interact closely and recently. In summary, the reader will found in next pages a multi-agent system based-on context and location for heterogeneous domain in ambient intelligence. We explain the detail of the user profile acquisition using genetic algorithms. And we present CALoR system to consolidate the relationship between agents. For each item, we expose our experiments and the obtained results, that validates this approach.
Researches on Ambient Intelligent (also known as Ubiquitous Computing) using wireless technologies have increased over the last few years. Ambient Intelligence is a new concept adopted for referencing an intelligent environment, where the user is assist by the computer and his mobile device wherever he is. Several technologies are necessaries in these environments to provide personalized services. All immersed devices are invisibles to the user. Ambient Intelligence is an ideal workspace for agents because of autonomous distributed and proactive agent nature. The main contribution of this Thesis consists of designing a methodological approach a Context Aware System (involving location services) using Agents that can be used in very different domains. We review several scenarios to define a multi-agent architecture, that support the information needs of these new technologies, that will be used in heterogeneous domain. We use ontology to facilitate the communication between agents. Furthermore, this ontology allows agent’s reasoning. In this manner, the use of geographic information allows to infer high level information about the real place where the agent is. In Ambient Intelligence scenarios, user preferences allow to select the best services for each user. This personalization is particularly suitable in a multi-agent system with learning capabilities. Based on this, and on the improvement of the mobile technologies, we developed an algorithm that allows learning the user profile. A list of prioritized services would therefore be presented to the user that takes into account user interests. Also, the providers agents could make their offers. Finally, users-providers interactions that take place in a dynamical domain needs a certain trust. Trust is build using trust and reputation systems. In this manner, we present CALoR reputation model that is a reputation model based-on context and location. This model allows the user agents of the systems to make recommendations from each other. CALoR take into account the past interactions of the agents and their spatial-temporal issues. We assume that a recommendation send by a user agent over a provider agent is more reliable when the agents interact closely and recently. In summary, the reader will found in next pages a multi-agent system based-on context and location for heterogeneous domain in ambient intelligence. We explain the detail of the user profile acquisition using genetic algorithms. And we present CALoR system to consolidate the relationship between agents. For each item, we expose our experiments and the obtained results, that validates this approach.
Description
Keywords
Inteligencia ambiental, Agentes inteligentes, Arquitectura multi-agente