This paper addresses the optimization of information retrieval in Arabic. Significant progress has been made during the past twenty years. The results derived from the expanding development of sites in Arabic, are often spectacular. Nevertheless, several observations indicate that the responses remain disappointing, particularly upon comparing users' requests and quality of responses. One of the problems encountered by the user is the loss of time when navigating between different URLs to find the desired responses. This may be due to the absence of forms morphologically related to the research keyword. Such forms are liable, in a number of cases, to be needed if the user is to obtain the answers he seeks. A second problem concerns the formulation of the query, which may prove ambiguous. This is frequently the case when the query word belongs to everyday language. We focus on contextual disambiguation based on a rich lexical resource including - among other things - collocations and set expressions. The overall scheme of such a resource, the completion of which is still to come, will be hinted at here. The need for such a lexical resource for the development of information retrieval in Arabic will emphasized. Our approach leads to the design of a multi-agent system. Our choice of an analysis system based on this approach is motivated by a need to solve some of the problems encountered when using conventional methods of analysis, and to improve the results of queries thanks to a better collaboration between different levels of analysis. We suggest resorting, at this stage of the research, to four agents, namely the morphological, lexical, contextual agents, in addition to a users' agent. Our agents will "negotiate" and "cooperate" throughout this analysis, starting from the submission of the request until the requested result is obtained.