This paper describes the architecture of an agent-based music service that generates dynamic playlists based on suggestions from multiple users.The system employs a number of agents capable of autonomy, mobility, scalability, collaboration and platform independence.In the system, agents are classified into two groups: requesters and music service team.Requesters work on behalf of their users, and bring music suggestions from the users to a DJ agent (DJ) which is part of the Tank service team.
The DJ has a reasoning component responsible for music selection and negotiation.Music selection is the result of negotiation among agents in sealed-bid auctions.The bidding strategy is specified by users.The amount of a bid implies the value of a song for a specific user; the DJ will evaluate this amount to determine the winner of the auction, whose song will be placed on the playlist.
All agents interact with each other using an interaction/negotiation protocol, through agent messaging.A proof of concept implementation of this here architecture shows that agents in the A-DJ system automatically perform tasks and collaborate with each other under the proposed protocol.The implementation appears to be scalable up to approximately 800 customers on a generalpurpose computer system.