I'm planning to perform sentiment analysis on reviews of product features (collected from Amazon dataset). I have extracted review text from the dataset and performed POS tagging on that. I'm able to extract NN/NNP as well. But my doubt is how do I come to know that extracted words classify as features of the products? I know there are classifiers in nltk but I don't know how I should use it for my project. I'm assuming there are 2 ways of finding whether the extracted word is a product feature or not. One is to compare with a bag of words and find out if my word exists in that. Doubt: How do I create/get bag of words? Second way is to implement some kind of apriori algorithm to find out frequently occurring words as features. I would like to know which method is good and how to go about implementing it. Some pointers to available softwares or code snippets would be helpful! Thanks!