An Efficient System to Predict Customers’ Satisfaction on Touristic Services Using ML and DL Approaches Said Gadri Laboratory of Informatics and its Applications of M’sila LIAM Department of Computer Science, Faculty of Mathematics and Informatics University Mohamed Boudiaf of M’sila, M’sila, 28000, Algeria said.kadri@univ-msila.dz Sara Ould Mehieddine Laboratory of Informatics and its Applications of M’sila LIAM Department of Computer Science, Faculty of Mathematics and Informatics University Mohamed Boudiaf of M’sila, M’sila, 28000, Algeria saraouldmahieddine@gmail.com Khadidja Herizi Laboratory of Informatics and its Applications of M’sila LIAM Department of Computer Science, Faculty of Mathematics and Informatics University Mohamed Boudiaf of M’sila, M’sila, 28000, Algeria herizikhadidja896@gmail.com Safia Chabira Laboratory of Informatics and its Applications of M’sila LIAM Department of Computer Science, Faculty of Mathematics and Informatics University Mohamed Boudiaf of M’sila, M’sila, 28000, Algeria safia.chabira@univ-msila.dz Abstract--- In the last decade, neural networks NNs become a favorable solution for many applications in artificial intelligence AI. For instance, the majority of tourism companies have professional websites where customers can book: flights, bus and taxi trips, hotels, restaurants, etc. they can also compare services in terms of prices, locations, services quality, and other interesting criterion. For this purpose, the used dataset consists of a sample of hotel reviews provided by customers who have reserved recently. Analyzing these reviews will help companies to know if their services are suitable for customers, satisfy their needs and what is the degree of this satisfaction. i.e., customers are happy or not? Satisfied or not? Our main objective in this work is to develop an efficient and intelligent system based on NNs which allows us to predict how customers feel about the provided services. To accomplish this work, we have proceeded to the classification task using many machine learning algorithms, including LDA, KNN, CART, NB, and SVM. Then, we proposed in the second stage a deep neural network DNN model to perform the same task. Finally, we established a short comparison between the different algorithms. In the programming stage, we benefited from the large opportunities offered by Python language, as well as Tensorflow and Keras libraries. Keywords: machine learning, deep learning, Artificial Neural Networks, Natural Language Processing, Social media; I. INTRODUCTION Today social media such as Twitter, Facebook, Instagram, Tiktok become an important means that allow people to share their opinions and sentiments about a product they want to buy or to express their views about a particular topic, company services, and political events [1]. Many companies and business organizations need to process these sentiments/opinions and exploit them in many interesting applications such as improving the quality of their services, drawing efficient business strategies, planning powerful and concurrent electoral campaigns, achieve a large number of new customers, and understanding public behavior for governments and states [2]. In our days, sentiment analysis is considered among the hottest research topic in NLP and text mining fields. It can be defined as the process of extracting automatically the relevant information that expresses the opinion of the user about a given topic [1, 3]. A simple form of such analysis would be to predict whether the opinion about something is positive, negative, or neutral (polarity). There exist other forms of sentiment analysis or opinion like predicting rating scale on product’s review, predicting polarity on different aspects of the product, detecting subjectivity and objectivity in sentences, etc [1, 2, 4]. Sentiment analysis is useful in a wide range of applications and domains, notably: business activities, government services, biomedicine, recommender systems. For instance, in the domain of business and e-commerce, companies can study customers’ feedback relative to a product in order to provide better: products, services, marketing strategies and to attract more and more customers [2]. In the field of recommender systems, we use sentiment analysis to improve recommendations for books, movies, hotels, restaurants, supermarkets, and many other services [5]. There exist four approaches to process the problem of sentiment analysis, including lexicon-based approach, machine learning approach, deep learning approach, and hybrid approach [1]. The lexicon-based approach was the first approach that has been used by researchers for the task of sentiment analysis. It is based on two principal techniques: the dictionary-based technique which is performed using a dictionary of terms like those in wordnet, and the corpus-based technique which is based on a statistical analysis of the content of docs combined with some statistical algorithms such as hidden Markov models HMMs [6], the Conditional Random Field CRF [7]. The machine learning approach [8] is proposed by many researchers for SA, and based on classical ML algorithms, such as NB [9], SVM [10], etc. Deep learning approach is recently proposed by researchers and know a large success in many fields, such as computer vision [11-15], image processing [16-19], object detection [20, 21, 22], transportation [23], network optimization [24], sensor networks [25 – 29], system security [30]. It gives better results in terms of accuracy but needs massive data. Many models are currently used, including DNN, CNN, RNN, LSTM. Our main objective in this work is to classify opinions expressed by customers by short reviews to determine whether the reviews’ sentiment towards the touristic service is positive or negative. For this purpose, we used the traditional machine learning ML approach and the deep learning DL