Abstract
Composition of Web services has received significant attention. While the business world has developed XML-based standards like WSDL for flow-based composition, the semantic web community focuses on reasoning about web resources by explicitly declaring preconditions and effects with terms defined in ontologies like RDF and OWL-S. However, most existing efforts assume Web services are passive computation entities that wait in registration centers for discovery and invocation. This paper proposes a paradigm shift toward autonomous "service agents" that proactively collaborate to fulfill user requirements. By integrating function ontologies with game-theoretic automated mechanism design and a multi-step negotiation framework, the authors demonstrate how decentralized agents can form stable, high-quality coalitions to solve complex tasks.
Introduction
Traditional Web service composition treats services as passive entities. This "search-and-invoke" model struggles with the dynamic, decentralized nature of the modern internet. To address this, the authors propose a framework for autonomous service collaboration that tackles three core challenges: establishing mutual understanding between independent agents, ensuring coalition stability among rational (self-interested) agents, and optimizing the balance between Quality of Service (QoS) and agent profit.
Methodology: A Three-Layered Approach
The proposed framework relies on three integrated components to enable requirement-driven collaboration:
1. Function Ontology
To enable mutual understanding, the authors define a formal terminology for describing resources, states, behaviors, and functions. This ontology allows agents to map their internal capabilities to specific state changes requested by a user (e.g., changing a database state from "closed" to "open").
As shown in the ontology, behaviors are defined by their ability to transition a resource from a starting state to an ending state. This state-based modeling is crucial for automated composition logic.
2. Automated Mechanism Design (AMD)
Drawing from game theory, the authors use AMD to establish "rules of the game" that ensure Individual Rationality (IR) and Incentive Compatibility (IC). By modeling the system as a Bayesian Nash equilibrium, they ensure that agents are incentivized to report their true capabilities and remain in a stable coalition rather than defecting for marginal gains.
3. Negotiation Framework
A three-step iterative process facilitates consensus:
- Proposal: The requester suggests an initial solution based on preferred QoS.
- Evaluation: Agents accept, refuse, or propose modifications based on their payoff requirements.
- Selection: The requester selects the best modified solution, repeating the cycle until consensus is reached.
Evaluation and Results
The framework was tested using a simulated numerical experiment in an E-Learning domain. The scenario involved creating and grading test papers using four types of agents: DatabaseSearcher, TestPaperCreater, InterfaceAgent, and Evaluater. Simulations across various agent/function configurations revealed:
- High Success Rate: Consensus was reached in 96% of cases for the 10-agent/20-function test group.
- Rapid Convergence: 97% of successful negotiations concluded in fewer than 10 rounds, demonstrating high computational efficiency.
- Mutual Benefit: Requesters achieved satisfaction degrees above 0.9, while agents frequently earned payoffs exceeding their expected averages.
Discussion and Future Work
The core innovation lies in treating Web services as BDI (Belief-Desire-Intention) agents, allowing for stability through mathematical incentives rather than assumed trust. However, a gap remains between this theoretical framework and industry standards. Future work aims to integrate this approach with BPEL (Business Process Execution Language) and WSDL to automatically generate executable composite services from formed coalitions. Additionally, the impact of real-world network latency on negotiation rounds in global-scale deployments remains to be investigated.
Summary updated