SENTIMENT ANALYSIS OF FACEBOOK DATA: A METHODOLOGICAL AND EMPIRICAL STUDY OF SOCIAL MEDIA OPINION MINING

Abstract

Social media has become an important source of information for understanding people's opinions, reactions and attitudes. Facebook, in particular, provides a large amount of textual communication through posts, comments and discussions. Unlike conventional survey responses, Facebook comments are generally spontaneous and are expressed in informal language. They may contain positive and negative expressions, neutral statements, sarcasm, abbreviations, emojis, spelling variations and, in multilingual societies, more than one language. These characteristics make Facebook sentiment analysis both useful and methodologically challenging.

The present study examines sentiment analysis of Facebook data from both methodological and empirical perspectives. The study reviews the development of sentiment analysis from lexicon-based techniques to traditional machine-learning, deep-learning and transformer-based approaches. For empirical illustration, a documented Facebook sentiment corpus containing 34,006 manually annotated comments is considered. The dataset contains 20,668 negative comments, 9,581 neutral comments and 3,779 positive comments. The distribution indicates a substantial predominance of negative comments, while positive comments constitute the smallest category. Such imbalance is important because it demonstrates why accuracy alone cannot be treated as a sufficient measure of model performance. The study therefore discusses precision, recall and F1-score alongside accuracy and recommends the use of stratified sampling, careful preprocessing and class-sensitive evaluation. The paper also examines the methodological problems associated with Facebook language, including sarcasm, negation, emojis, spelling variation and multilingual communication. The study concludes that successful Facebook sentiment analysis depends not only on the choice of classification algorithm but also on the quality of the dataset, annotation procedure, preprocessing, class distribution and evaluation framework.