On the Analysis of Joining Communities of Agent-based Web Services (Extended Abstract) Jamal Bentahar Concordia University, Canada bentahar@ciise.concordia.ca Babak Khosravifar McGill University, Canada babak.khosravifar@gmail.com Kathleen Clacens, Christophe Goffart, Philippe Thiran University of Namur, Belgium kclacens,cgoffart,pthiran@fundp.ac.be ABSTRACT Communities of agent-based web services are virtual groups gath- ering functionally equivalent web services having different non- functional attributes. Building reputable communities hosting re- liable web services is still an open challenge. In this paper, we pro- pose a mechanism that web services through associated agents can use to join existing communities. Key components of this mech- anism are agents providing information about potential members of communities. Analyzing incentives for these agents to reveal accurate information is the main contribution of this paper. Categories and Subject Descriptors I.2.11 [Distributed Artificial Intelligence]: Multiagent Systems; J.4 [Computer Applications]: Social and Behavioral Sciences - Economics Keywords Communities of web services; game theory 1. INTRODUCTION Recently, some research proposals have demonstrated the im- portance of grouping functionally equivalent web services within communities providing these web services with high visibility and many security and management advantages [2, 3]. In this context, selecting reliable web services to be part of a given community is a challenging issue that still needs further consideration. We abstract web services as autonomous intelligent entities, which are benefit maximizers in the sense that their objective is to get a maximum number of requests. The contribution of this paper is a game-theoretic model ana- lyzing the communities of agent-based web services from the per- spective of hosting different web services. Web services initially act alone and analyze the benefits of joining a community. A game is set between the master web service acting as the manager of the community and services acting as information providers within a group called information service group. Agents in this group, called information services, provide the necessary information re- garding the web service that is attempting to join a community. The Appears in: Proceedings of the 12th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2013), Ito, Jonker, Gini, and Shehory (eds.), May, 6–10, 2013, Saint Paul, Minnesota, USA. Copyright c  2013, International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved. involved information services can either lie or tell the truth about the requested information. It is worth mentioning that web services themselves could be malicious and collusion between web services aiming to join a community and information service agents should be considered. In this paper, we analyze this important issue be- cause information services can be incentivized from web services to exaggerate about their qualities or even provide bad recommen- dations about other competitive services [4]. In [1], the authors addressed the problem of extracting truthful opinions from large groups of agents that could be service agents in online feedback systems by 1) designing various incentive-compatible payments and rewards; and 2) addressing the problem of collusion. However, this model addresses the reputation in environments where web ser- vices and agents function alone with no plan of cooperation. 2. PAYMENT FUNCTION The utility u k (x) an information service agent k aims to maxi- mize by choosing the strategy x ∈X is a function having 3 incen- tive components where the service agent obtains rewards or penal- ties according to the chosen strategy and considering the truthful probability of the master agent which provides the payment. The master’s truthful probability is modeled here using Beta-mixture distribution because data on the service’s behavior is not fully avail- able. The truthful probability of the master is computed using a combination of L parameters: p(D)= L  l=1 π l p l (D|θ l ) (1) where D =[O1,...,ON ] is an N dimensional vector of obser- vations on the master agent’s behavior, p l is the l th parameter dis- tribution and θ l and π l are the two distribution parameters, which are usually estimated using the maximum-likelihood. Particularly, π l (l ∈ L) is the mixing coefficient that controls the contribution of each trust parameter in the overall trust value. The problem the information service agent should solve is: x * = argmax x∈X u k (x) where u k (x) is defined as follows: u k (x)= p(D)(f k (x)+ g k (x)+ c.h k (x)) (2) The first component f k (x) is a positive reward that a customer gives to the information service agent k that is willing to provide the asked information. The second incentive component g k (x) cor- responds to a value that will be granted to the information service 1339