ADAM: ADvisory Agents Modeling System to Enhance Student-Supervisor Decision Making Maedeh Mosharraf, Fattaneh Taghiyareh, Atiyeh Soleimani, Fatemeh Orooji Electrical and Computer Engineering department, University of Tehran Tehran, Iran {m.mosharraf, ftaghiyar, at.soleimani, f.orooji}@ut.ac.ir Abstract— Web rapid development has provided new learning environments, bringing online education as a necessity in many sectors of the society. In such an environment, selecting supervisor is a critical decision that graduate students as well as professors are involved with which could benefit from e-learning tools. In this paper we have proposed ADAM, an ADvisor Agent Modeling system, which is a solution for supervisor selection based on multi-agent. In order to simulate student-professor relation, we have profiled them through deriving their decision parameters and other required information, using data obtained from various sources including Learning Management System (LMS), Community of Practice (COP), as well as our question answering user interface. ADAM utilizes a weighted ontology, which is proposed by authors, to improve agent decision making based on their provided model. Proposed advisory system is implemented at the University of Tehran, using more than 50 different profiles of students and professors. ADAM can be embedded in any web based educational system and any group activities which need forming hierarchical relationships between different members who have their own demands and goals. Keywords— Learning environment; Multi-agent; Advisory system; User modeling; Weighted ontology I. Introduction In e-learning, multi-agent based assistant systems appear to be a promising approach to deal with the challenges in educational environments. They can provide new patterns of learning and applications, such as personal assistants, user guides and alternative help systems, which are helpful for both students and teachers in their computer-aided learning-teaching process [1]. Various academic activities are applicant through some multi-agent assistant systems each use special mechanism for agents decision making. For example [2] has introduced a voting based system for students’ course selection where intelligent software agents allocate points to different courses on the behalf of the students and voting occurs in several rounds. In this system agents utilize student preferences and information provided from previous rounds to vote intelligently and strategically. [3] has proposed an assistant system to recommend a personalized learning path for each learner based on his/her profile, contents features and the specifications of the learning domain. It has presented an ontology base semantic view to overcome the problem of finding appropriate learning contents from a wide range of course contents in order to refine recommended contents after concepts relevance calculation. In [4] a multi-agent system is proposed that helps learners to enhance self-monitoring, self- evaluations and self-regulation. Furthermore an interactive teaching agent which contains hints, messages and immediate questions and answers is designed to support it. [5] has presented a tree similarity algorithm for matching agents in commercial environments where agents’ negotiations mechanisms can be applied to e-learning environment interactions between course provider and learners. All agents have their own semantic trees which describe their context of activity, and similarity between them is calculated based on average weights. Using agents as human assistances could lead to promising results, if agents can perceive user needs and attitudes perfectly which require an effective user modeling. Creating and managing user models need some techniques to integrate distributed information about different users in a central server as it proposed in knowledge tree[6] or building various models of a user representing his/her profile related to different domains. This requires some mechanisms for knowledge transferring between different domain specific ontologies as it is introduced in [7]. Some of common metrics for human-robot interaction are: understanding user information and needs, modeling the information well, measuring user feedbacks and applying his/her view to the model [8]. User modeling can be implicated through an intelligent interface that tries to apply user needs and personal characteristics to agent’s behaviors [9] which can