Sentiment Analysis Framework Using Data Driven Approach

The Internet is a free and straightforward way to access an immense measure of crude content information that can be mined for sentiment analysis. For a long time, sentiment analysis has been used for market research, user opinion mining, recommendation systems, and analysis of people’s views on a topic. Many researchers have developed techniques for sentiment analysis, yet many complications remain. Selecting and understanding attribute patterns in a text dataset is essential to build a good model and know where this model can be used. Different text datasets have different relations between their attributes and classes. For example, let us take a dataset with totally random English texts labeled as positive or negative. We expect that extracted attributes for the positive or negative class are very heavy with general words that we consider positive or negative in everyday English use. However, if the dataset is created on a niche topic, such as an economic pandemic, we would probably see that positive and negative classes are heavy with words specific to these topics, or they may not be considered the classifier. However, we might want to give importance to those niche-specific attributes specifically. In this paper, we take five different datasets of different instance lengths. We go through some attribute selection techniques and use them under some classifiers to visualize a pattern, do sentence-level sentiment analysis, and finally extract patterns from the datasets to analyze them. There are few related works on these datasets, and our technique performs better than the existing works. This paper aims to present a method that can easily be fruitful for text mining and decent accuracy to any dataset.