Social networks have become a valuable platform for tracking and analyzing Internet users' feelings. This analysis provides crucial information for decision-making in various areas, such as politics and marketing. In addition to this challenge and our interest in the field of big data and sentiment analysis in social networks, we have dedicated this work to combine different aspects of methods or techniques leading to the facilitation of feelings classification in social networks, including text analysis and sentiment analysis. We expose the approaches and the algorithms of supervised machine learning for the classification of feelings. We further our research to concisely present the methods of data representation and the parameters used to evaluate a sentiment analysis method in the context of social networks, with a section presenting our novel lexicon-based approach to give more accurate results in classifying Arabic text. The proposed approach has shown a promising accuracy percentage, especially the precision of the sentiment detected from text with F-Score up to 66%.