Conversational AI for companionship

Abstract

The conversational style of a human is estimated by humor, personality, voice tone, etc. These characteristics are necessary for virtual assistants that are artificially intelligent for conversation. This research recommends an intelligent system capable of holding an appropriate human-like dialogue, including the emotion and personality of a specific character. To draw the pattern of the attributes of specified emotion, a method can be used to transmit voice tone. In order to determine all the necessary characteristics mentioned above, the goal is to use different categories of machine learning models. Since the pattern of conversation, style varies from one individual to another and geographically, our goal is to create a virtual assistant. In addition, a conversational model will be applied to it. It will read the category of emotions(exclamation, assertion, negation, interrogation) of human beings and respond accordingly. Many methodologies are being utilized to predict sentiments through AI and react accordingly. IVA is one of them but with its limitations and boundaries. Therefore, this paper comes with several methodologies that can be used alongside IVA; such as HMM, GMM, SVM, NLU, BoAW, BERT, etc. These algorithms and methodologies will help to predict the sentiments used in a context and precisely predict the outcome of an inquiry. To sum it up, this thesis aims to create a conversational AI for companionship, which will create an emotional bridge between itself and the user.