Fuzzy Prediction Model to Measure Chatbot Quality of Service

Abstract

Detecting breakdown is a common phenomenon in the conversational system, which is referred to when the system fails to provide appropriate responses to the user. Existing studies are detect breakdown using different features such as word similarity, topic transition, and clustering. In this paper, we focus on the different important feature, which is human thinking and reasoning. We use this feature to model chatbot quality of services (CQoS) based on detecting the breakdown. Thus we introduce the fuzzy prediction rule-based framework to measure chatbot quality of service by detecting the breakdown utterance considering end-user and chatbot points of view. Inputs utilized in the proposed fuzzy logic-based model are multiple useful features extracted from utterances. The outputs are the degrees of relevance for each utterance to the quality of services. Several fuzzy rules are designed, and the defuzzification method is used in order to achieve desired CQoS results. Based on the outputs from the fuzzy model, the handover mechanism will activate. We evaluate the proposed formwork with other state-of-the-art models.