On Some Learning Methods among Cooperative Agents

Abstract

The characterizations of user' activities and possibly their customization according to the user's needs are an important issue in the design and maintenance of websites. Server weblog files have rich information about user's current interests, which can be mined. This mining analysis generates a tool to guide the users in their browsing activity and assist them to obtain more relevant information. The thesis specializes in attempting to treat the problem of navigating a Website and retrieve information desired by a visitor as soon as possible. In this thesis, a new methodology based on some learning techniques for a Web-Based Multi-agent application is suggested. The methodology attempts to discover the hidden patterns in the user's visited links. It presents a new approach that involves unsupervised, reinforcement learning, and cooperation between agents. This approach is utilized in an experimental website to discover patterns that present in a user's profile and classify them into specific categories of materials using significance percentages. The profile provides recommendations of interesting categories and links to a specific user. The experimental results of the suggested approach exhibited a successful user pattern recognition and cooperative learning among agents to obtain enhanced user profile. This indicates that combining different learning algorithms may improve user satisfaction indicated by some metrics such as the percentage of precision, recall, the progressive category weight and F1-measure.