User v/s bot : analyzing sentiment from legal chats

Abstract

Sentiment analysis, a subfield of Natural Language Processing used to mingle its arms with Machine Learning (ML) is becoming popular in recent trends as we have huge amount of data in text, audio or video in WWW. In global spectrum, chatbots are becoming the essential assistants in our daily activities and the developments of such bots are being carried out under the umbrella of discourse and pragmatic sections of NLP. The intelligence is added and fuelled into the bots while it tries to determine emotions from such textual or voice data. The goal of sentiment analysis is to define automatic tools able to extract subjective information, like opinions and sentiments from language texts and pictures. One of our objectives is to study sentiment from chat data from the perspectives of both user and bots so that to shape the response of the bots based on user sentiments and vice versa. On the other hand, it’s one in every of the key chatbot features is to analyze customer data by mining thoughts, opinions, or sentiments. The employment of conversational sentiment analysis enables a chatbot to grasp the mood of the customer by sentence structures and verbal cues. Bots can use sentiment analysis to switch responses in tune with customer’s emotions and thus help segment the audience. User-bot is an interactive conversational agent. User-bots are employed in different domains like gaming, customer service, information provider etc. Siri, Alexa, Cortana (given as examples for such conversational agents). Sentiment analysis from natural language texts is the process of identifying subjective clues and categorizing the text or statement into positive, negative or neutral classes. The present work describes about incorporating sentiment analysis on User-bot interactive dataset. Our goal is to review dialogues of user and bot datasets and find out the sentiments and analyze those sentiments. We have used supervised as well as deep learning models to analyze the sentiments. By using Logistic Regression method, we have achieved 77% accuracy and using RNN and LSTM models, we have obtained 80% and 85% accuracies, respectively