The instinctive medium that humans use for communication is of words and not numbers. Ratings and reviews are the most common form of feedback that customers of a product or service can provide on an online platform. While ratings are quantitative, reviews are expressive. Extracting the users’ true sentiments from their review with respect to each aspect is highly insightful. Our project leverages the Stanford CoreNLP Parser to apply PoS tagging, Coreference Resolution and Dependency Inferencing for constructing aspectsentiment pairs. The aspect polarities are calculated using an amalgam of sentiment lexicons like VADER and TextBlob. We have also used some filtering rules with hard limits to ensure that our system only has the most relevant reviews for processing. One filter obstructs spammed reviews and blacklists the reviewers for the same. Another makes sure that the reviews to be processed have been marked helpful by a strong majority of users. Our system provides recommendations to users based on prioritised aspects. We construct user profiles and product profiles to map according to aspect preferences. The Stanford CoreNLP Parser dependencies have been thoroughly exploited to design the rules for aspect-sentiment extraction
Keywords : Natural Language Processing, Sentiment analysis, Opinion parsing.