Conquest: um framework para a construção de chatbots de IQA baseados em templates sobre knowledges graphs

Abstract

In this dissertation CONQUEST is presented, a framework that automates much of the construction process of chatbots for the task of Template-based Interactive Question Answering (TBIQA) for Closed Domain Knowledge Graphs (KG). The chatbots produced by CONQUEST are able to answer questions issued in natural language, using conversation to solve the possible problems of ambiguity and the lack of information needed to formulate the answer. For the interpretation of the question, the system performs the process of classifying the user intention to one of the templates defined by the developer a priori. For this, CONQUEST has a flexible question classification mechanism based on Machine learning (ML) that uses both textual features and semantic features and that adapts to use, learning new ways how the same question can be realized. This classification mechanism makes a chatbot capable of dealing with the problem of linguistic variability. In addition, a chatbot produced by CONQUEST uses the user feedback in its training, adapting to the use. As main contributions of this work, we have: (1) A new architecture for chatbots for gls TBIQA and (2) A tool for the automatic building of chatbots for TBIQA for KGs. Finally, with CONQUEST, the developer invests his time only in creating the question templates supported by the system, leaving the control of messages, natural language processing, question interpretation, querying data sources and generating responses for the framework deal with it.