Opinion mining in social media

Nowadays, social networking sites like Twitter, Facebook, YouTube have gained so much popularity and they had become a significant part of our daily life. People tend to share their ideas on social media platforms and massive amounts of data are generated daily. The mining of sentiments expressed in huge opinionated text data has become an increasingly popular research field. The unstructured and huge volume of social media posts makes the data extremely challenging to analyse with high accuracy and efficiency. In previous researches done, most of the existing computational models for classifying sentiments from informal documents relied heavily on machine learning techniques. Feature selection is the key step in training the sentiment classifier with machine learning techniques, however, there is no definite conclusion on how to choose the features that can enhance the performance of classifiers in terms of accuracy and computation time. In this work, experiments were conducted using both machine-learning-based approaches and hybrid approaches to explore the importance of feature selection in sentiment analysis. A near-linear relationship had been observed between log(no_of_features) and classifier accuracy. When the accuracy exceeded the maximum point, a slowly decreasing trend was observed. Results obtained were validated and similar trends existed in both binary and multiclass classification. Hybrid approaches had also been applied and the accuracies were compared with machine- learning-based methods. The results of the comparison further proved that the performance of a classifier could be enhanced with proper feature selection.Bachelor of Engineering (Electrical and Electronic Engineering