Towards Usage Policies for Fuzzy Inference Methodologies for Trust and QoS Assessment

Abstract

In this paper, we discuss the benefits of several fuzzy inference system design methodologies and evaluate their characteristics in regard to our trustworthiness and QoS measurement models. Our analysis shows that Mamdani-Assilian or Larsen type and Takagi-Sugeno-Kang type fuzzy inference methods have their merits in different situations. We propose to equip an autonomous agent which acts on behalf of a human being with a policy table enabling the agent to dynamically decide which fuzzy inference system it will select during the trustworthiness evaluation process. We argue that in most situations the Mamdani-Assilian or Larsen type fuzzy inference system represents the preferred choice. However, in situations where the fuzzy rulebase is large, the Takagi-Sugeno-Kang type fuzzy inference system should be chosen due to its superior performance characteristics. This way the agent can perform its tasks more efficiently by choosing the appropriate calculation method depending on the given circumstances. The assessment of trust and credibility is part of our daily life it happens subconsciously and is based on recommendations, past experiences and vague feelings. Reliable and precise measurement of trust and credibility is especially important if we want to achieve autonomous interactions of intelligent agents in unsupervised distributed environments. However, the replication of such social behavior in information systems represents a major challenge. In an ideal scenario, a person, who wants to purchase goods or consume a service, would instruct his intelligent agent to execute this time-consuming task on his behalf. The agent's duties and responsibilities would involve service discovery, service selection, contract negotiations, service execution or consumption, payments, and reviewing of the delivered service quality. In previous work, we have proposed models for both, trustworthiness evaluation in distributed environments to support selection of potential services [1], as well as a quality of service (QoS) review model [2]. Both models are based on fuzzy logic [3] which offers a mathematical concept to deal with uncertainty for the calculation of outputs. This ability to offer reasoning capabilities based on uncertain or incomplete information makes it suitable to simulate human reasoning which is based on similar principles. In both models, we have chosen the Mamdani-Assilian (MA) [4] approach for the fuzzy inference process of our models. In separate research, our group has also proposed a fuzzy model based on the Takagi-Sugeno-Kang (TSK) [5] inference method to determine the trustworthiness and credibility of peer agents in distributed environments [6]. In this paper we discuss the benefits of these and other fuzzy inference system (FIS) design methodologies and evaluate their characteristics in regard to our trustworthiness and QoS measurement models. First, we briefly introduce our fuzzy trust and QoS evaluation models to establish the context for our suitability analysis. Second, we provide details on the fuzzy inference methods before comparing their benefits in the different situations that the agent might encounter. Based on this analysis we will finally introduce a policy based model which assists the agent to select the appropriate FIS for the specific situation the agent encounters.