A task model for cooperative information gathering agents

摘要

Cooperative Information Gathering (CIG) considers information retrieval as a distributed problem solving activity carried out by intelligent and cooperative Agents. A CIG system may be abstracted as three related models: an informational model (IM), an organizational model (OM) and a task model (TM). This paper focuses on the task model which describes the information gathering process: the recursive decomposition of a user-task into sub-tasks, the allocation mechanism of sub-tasks to agents, the cooperation between agents and with the user. This paper shows that current Task Representation Languages do not adequately model CIG tasks and notably the following features: concurrency, cooperation, openness, different level of abstractions. In addition, some existing languages define tasks at a low level of detail, as the process to be executed, and do not provide abstractions to design incrementally the task model. Conversely, other models define tasks at a very high level, as a goal to be reached, and do not provide an operational semantics. Our approach stays between these extremes. At a conceptual level, we have extended the And/Or Graph formalism to take into account the particularities of the information gathering process, and to provide the designer with an easy-to-understand and accurate task representation language. We also have defined an algorithm to automatically derive a Petri Net with Objects from this extended And/Or graph. The Petri Net with Objects formalism captures all the patterns of the conceptual model, provides an executable specification, enables simulations of tasks executions and proofs of behavioral properties. The main advantage of our approach is to ease the work of the designer and to provide an automatic mapping from specification to execution.