Cost-Sensitive Command and Control for Web Content Surveillance Using Autonomous Agents

In this paper, a cost-sensitive multi-agent approach for the command and control of web content surveillance is selected and its key functions are fully described; Based on these functions, a formal expression for cost-sensitive command and control is provided. The structure of a C4ISR system based on autonomous web surveillance agents is also presented, in which the intelligent components of text processing, image processing, and topical crawling are used. For the image processing component, a cost-sensitive learning method for deep convolutional neural networks and a method for cost-sensitive topical crawling have been proposed. In the proposed image processing method, the process is stopped when the middle classifiers, which have been added to the conventional CNN structure, reach the certainty necessary to determine the class of a sample; otherwise, the classification continues in the higher layers of the network. In this way, processing resources are managed and costs are reduced. In the proposed topical crawling method, instead of using only one link extraction method, the scores of a set of link extraction methods is used to increase the efficiency, which leads to the targeted use of bandwidth. The results of the experiments show the efficiency of the proposed methods in comparison with the stage of the art results. The cost-sensitive approach presented in this article, in addition to web content surveillance, can be used for command and control of other real-world problems.