Bulletin of Electrical Engineering and Informatics Vol. 13, No. 3, June 2024, pp. 1752∼1761 ISSN: 2302-9285, DOI: 10.11591/eei.v13i3.6966 ❒ 1752 Empowering customer satisfaction chatbot using deep learning and sentiment analysis Abdelhak Merizig 1 , Houcine Belouaar 1 , Mohamed Mghezzi Bakhouche 2 , Okba Kazar 3 1 LINFI Laboratory, Department of Computer Science, Mohamed Khider University of Biskra, Biskra, Algeria 2 Department of Computer Science, Mohamed Khider University of Biskra, Biskra, Algeria 3 College of Computing and Informatics, Department of Computer Science, University of Sharjah, College of Arts, Sciences Information Technology, University of Kalba, Sharjah, United Arab Emirates Article Info Article history: Received Jun 12, 2023 Revised Sep 27, 2023 Accepted Oct 24, 2023 Keywords: Artificial intelligence Chatbot Deep learning Human computer interaction Natural language processing Sarcasm detection Sentiment analysis ABSTRACT The rapid advancement of technology holds great promise for various types of users, clients, or service providers. Intelligent robots, whether virtual or physi- cal, can simplify the reservation process. With the development of machines and processing tools, natural language processing (NLP) and natural language un- derstanding (NLU) have emerged to help people comprehend spoken language through machines. In order to facilitate seamless human-machine interaction, we aim to address customer needs through a chatbot. The objective of this paper is to incorporate sentiment analysis techniques with deep learning algorithms to cater to customers’ needs during message exchanges. This study aims to create an intelligent chatbot to engage customers during their routine operations and offer support. In addition, it offers to companies a manner to detect sarcastic messages. The proposed chatbot utilizes deep learning techniques to predict users’ intentions based on the questions asked and provide a helpful and con- venient answer. A new chatbot for the customer is presented to overcome with challenges related to a wrong statement like sarcastic one and feedback towards user messages. A comparison between deep and transfer learning gives a new insight to include sentiments and sarcasm detection in the conversion process. This is an open access article under the CC BY-SA license. Corresponding Author: Abdelhak Merizig LINFI Laboratory, Department of Computer Science, Mohamed Khider University of Biskra Biskra, Algeria Email: a.merizig@univ-biskra.dz 1. INTRODUCTION The evolving technologies in artificial intelligence are considered the primary helpful tool in human life [1]. The emergence of intelligent systems and their flexibility has facilitated their incorporation into web- sites and mobile applications. Human-machine interaction made a huge splash recently, especially during and after the COVID-19 pandemic, which introduced virtual and physical bots [2]. To prevent any direct commu- nication or even travel to book a restaurant or flight, usually people or clients must give a call or go directly to finish the job. Evolutionary techniques, such as natural language processing (NLP) and deep learning tech- niques, make it easier in a manner that makes machines ensure the communication operation. Virtual assistants have been known for many years through some applications such as Apple Siri or Amazon Alexa, and others make good help for users [3], [4]. Indeed, chatbots are considered the most suitable application to facilitate machine and human communication. The revolution of social media, such as Web 3.0 and Web 4.0, spread Journal homepage: http://beei.org