Comparative sentiment analysis of selected car models for an automotive client

Overview
A multinational automotive manufacturing company wanted to analyse the customer’s views on selected models vs the competitors
Solution
Data extraction
- Extracted data from the Facebook pages such as follower base of the page, number of posts, like and comments on each post, type of comments.
- Extracted data from car review websites such as cardekho.com. Fields extracted such as overall rating, likes, number of reviews, and comments.
Data cleaning
- Translate non English posts using Google API
- Perform stemming and lemmatization
- Remove punctuations, stop words
- Correct misspelled words using NLP
Data analytics and modelling
- Calculate sentiment score for each comment
- Topic, sentiment and car model wise review topic modelling
Data visualization
- Model comparison metrics developed on sentiments, topics and cars
- Social media presence comparison
- Word Cloud

Impacts
More than 50% users have negative view of car X
Car Y has significantly higher positive reviews percentage
Car Z needs to catch-up with its competitors
Car X has significantly high social media presence
Car Y is organically catching up as its review count is quite high compared to page likes and follow count