Conversational AI as Question Answering system implementation using BERT

Conversational AI refers the interaction of users throughmessaging apps or virtual assistants. Such communicationsthrough virtual agents or chat bots are designed to provide personalizeduser experience. Chat bot interactions are driven bylong tail messages, queries or two-way interaction with the privateaudiences. Real power of conversational AI is in its abilityto provide highly personalized interactions with huge numberof customers all together. As a business use case, conversationalAI can transform traditional ways of communication by facilitatingdepth engagement with users. One powerful applicationof conversational AI is question answering system. Either itwould be a virtual assistant or chat bot, or any support systemdesigned by any company, question-answering model plays acrucial role to provide such services. An automated questionanswering system enable a venture to provide its users a moredynamic and personalized experience. Practically, it’s difficultfor a human being to answer all the queries asked by users atone time. If anyone wants to do so, more resources will beneeded to be engaged with customers to solve their issues inservices or products offered. Automated question answeringsystem, not only saves a lot of time but also it acts efficienttool to understand customer needs and their behaviour. GeneralArchitecture of question answering process is to take inputfrom users, query analysis, information retrieval from the database, extraction of the most relevant answer and then outputthe desired answer. It is not as easy as it seems. Why? It is dueto the several challenges faced during the modelling. In thissession, we will discuss several challenges faced to design aquestion answering system like data availability, quality of thedata, it’s interpretability and language barrier. Though servalalgorithms have been implanted to design most efficient questionanswering system, BERT outperform all. We will discusshow Bi-directional Encoder Representation from Transformer[2] algorithm helps in solving mentioned challenges and providean optional solution. This tutorial will focus on designingquestion answering model by using human’s favourite tool forcommunication: Natural language processing. For this session,I will assume familiarity with basic terms of natural languageprocessing and python. The body of the talk will focus on implementationof BERT for designing question answering model,feature extraction techniques for data.