Master Thesis - 2009

Recently, as networks widely spread, the number of contents over networks is increasing. For efficient and flexible information retrieval over the such networks, agent technology receives much attention. There is the case where a retrieved information has difference of quality by user’s request and can be given scores. We proposed an agent execution control method for time-constrained information retrieval. It achieves to find better results by termination of an agent which already acquires results with enough quality or has less probability to improve the quality by continuing retrieving. In this method, however, all users are supposed to request just one result with the highest score. It is not realistic. In addition, this method ignores active cloned agents in the network. In this paper, we enhance the method to allow users to request multiple results. Moreover, it perceives progress reports from cloned agents for more accurate control. I introduce the expected improvement of the total score of requested results as a new measure for the execution control. It is developed by reports from all nodes where cloned agents run. Based on this measure, an agent which has less probability to improve the measure is terminated. Finally, the performance of the proposed method is evaluated by simulation experiments. It is confirmed that the enhanced agent control method improves the total score of requested results. Moreover, it keeps the fairness between users even if the numbers of requested results are not identical.