Mood Detection Based on Arabic Text Documents using Machine Learning Methods

Abstract

Document text classification is utilized for information feature extraction and retrieval as the primary source of digitizing the written information using text classification techniques. Text classification can provide much more information by analyzing the text using machine learning methods. One of the practical applications of text classification is mood detection using machine learning algorithms. Machine learning algorithms allow a practical and beneficial platform for analyzing and detect mood from the text documents. However, there are few applications to analyze the text in Arabic with high accuracy and especially detecting mood using Arabic text documents, messages, or blogs. The main objective of using machine learning algorithms is to detect the accurate mood target value to the given messages. This study focuses on four mood classes (Happy, Sad, Angry, fear). The text analyzed for this study were gathered from some social media and internet blogs. Three kinds of techniques have been identified based on machine learning approaches: Naïve Bays algorithm, k-Nearest Neighbors (KNN) algorithm, and Support Vector Machine (SVM) algorithm. The text was further analyzed for feature selection related to text mining, feature correlation analysis, and information gain. Lastly, splitting the text into training and testing sets for possible models using robust classifiers. After running the selected classifiers for our study, the results showed that Naïve bays classier had the highest achievement in terms of accuracy. Naïve Bays classifiers received 70%, support vector machine classifiers obtained 68.33%, and k-Nearest Neighbors (KNN) algorithm yield 51.67%.