Abstract:
Sentiment classification of product reviews data is mostly associated with assigning the review sentence with an overall opinion score i.e. positive negative or even sometimes neutral. This technique does not account towards the fact that people discuss various features of a product in a review and once someone talks about one positive and one neutral feature, the review sentence gets assigned a neutral score. This project is about addressing this limitation, by doing sentiment analysis on aspect level. We have two major tasks. One, aspect identification. Two, assigning opinion orientation to the aspect term identified. For the first task we use association mining rule presented by Hu and Liu [4]. Using this mining rule, we find the most recurring nouns and noun phrases which are our aspect terms. For the second task we used VADER [3], a sentence level sentiment classifier available in NLTK(Python) library. We developed a technique to identify the sentiment orientation of an aspect term in a sentence. This project helps understand the barriers and opportunities for further development in the field of aspect-based sentiment analysis.
