A feature-based trust sequence classification algorithm

Trust is a paramount factor in the development of service-based communities, where services continuously collaborate to successfully perform their tasks. Trust assessment helps users and services identify which partners to interact with. We tackle in this paper trust from an objective data mining perspective. We propose a novel feature-based approach to assess the trust behavior of a service. A trust behavior is represented as a sequence of trust observations during a certain time frame. By analyzing the possible trust behaviors of services, trust patterns are defined to describe trust sequences based on three criteria: its overall behavior, the starting behavior and ending behavior. Our approach spans over a rule based Prefix-Suffix Algorithm (PSA) for the classification of trust sequences. PSA computes new attributes to capture the chronological and structural nature of trust. Following a divide and conquer strategy, the trust sequence is divided into two parts each classified independently. PSA leverages some predefined merging rules to derive the class of the whole trust sequence from the classification results of these parts. We show the efficiency and accuracy of our approach by analytical and experimental evaluation.