A Technical Survey on the Modeling of Topical Bot

Aishna Gupta,Anuska Rakshit

doi:10.11648/j.ajsea.20211001.12

Abstract

This paper explains the working of the conversational AI models and their characteristic features. The primary objective of this paper is to let the readers know about what topical chats are and how they work. Topical chats have huge data/knowledge stored in them for making the conversation interactive and engaging with humans. The first-generation conversational AI models were simply focused on short task-oriented dialogs, such as telling jokes, the weather of the day, or playing songs. But now advanced models can have everyday smooth conversations. These models are built to understand different languages and their different accents. These models can identify whether the user is female/male/other, detect the change in the user’s emotion during the conversation and switch the topic of discussion accordingly. Building a conversational AI model has been a challenging task for researchers as well as the developers as they require deep knowledge in NLU, ASR, LM, Semantics, etc. Understanding human emotions and sentiments is a difficult task for an AI model. Recognizing the speech and giving a sensible response is challenging too. But nowadays AI models are so developed that they can even differentiate between good words and slang words.

Highlights

Building a conversational AI model has been a challenging task for researchers as well as developers
Challenging things for a conversational AI model are correlating with human emotions and sentiments, understanding the gender by pitch or tone, picking up the right search result for the user and for researchers having deep knowledge of NLU (Natural Language Understanding), ASR (Conversational Automatic Speech Recognition), Inappropriate Response Filtering and NLG (Natural Language Generation) is a crucial task
SPEAKER GENDER ANALYSIS Gender detection systems based on Gaussian Mixture Models, i-vectors and Convolutional Neural Networks (CNN) were trained using an internal database of 2,284 French speakers and evaluated using REPERE challenge corpus out of which the convolutional neural networks (CNN) system obtained the best performance with a frame-level gender detection F-measure of 96.52 and a hourly women speaking time percentage error below 0.6% [15]

Summary

Introduction

Building a conversational AI model has been a challenging task for researchers as well as developers. Researchers faced problems in image recognition and speech recognition. Voice assistants like Siri, Alexa were made, to achieve the goal of having a smooth conversion with humans. They are highly advanced and successfully deployed devices. Challenging things for a conversational AI model are correlating with human emotions and sentiments, understanding the gender by pitch or tone, picking up the right search result for the user and for researchers having deep knowledge of NLU (Natural Language Understanding), ASR (Conversational Automatic Speech Recognition), Inappropriate Response Filtering and NLG (Natural Language Generation) is a crucial task. This paper elaborates on the working of the topical chat and the challenges faced during the entire process

Conversational AI

Science

CONVERSATIONAL DATASETS AND COMMONSENSE REASONING

INNOVATION AND FEEDBACK FRAMEWORK

Findings

Conclusion