This repo contains a Python script that will go through the user reviews from an app and generate insight for the app creator. This is done by categorising each review into "positive", "negative", or "neutral". The insight will contain the distribution of reviews across these 3 categories so the creator knows how well the app is doing at a glance without having to look through all the reviews themself.
AI Sentiment Analysis The core functionality of this project is derived from an AI process called Sentiment Analysis. There are 2 main ways of analysing the sentiment behind natural language with different strengths: NLP (Natural Language Processing) It's possible to use NLP methods to understand better the meaning of text written in a natural language like English. This allows for a better understanding of the sentiment and strength behind a review than the other methods. However, it needs to be extensively trained and programmed which requires lots of data and computational power. This would only be feasible for this project if I outsourced it to another company and used their API (which would be very slow and dependent on their API functioning). Lexical Word-Based sentiment analysis syntactically breaks down a sentence and identifies key tokens influencing its decision for a particular sentiment. More complex algorithms also deal with negation and get more accurate. The benefit of this is that it's programmed, not trained. So it uses much less computing power can all be done locally and provides fast results.
For those reasons, I have sacrificed some accuracy for speed and computability and gone with a lexical sentiment analysis approach for this project.
METADATA The dataset is a comma-separated document but imported to Python as a text file. Each line is a separate review. Within each line, there will be 2 values separated by a comma. The 1st value is the name of the app. The 2nd value is the actual review.