Twitter Sentiment Analysis of US Airlines

Approach for Sentiment Analysis

• VADER Lexicon
• TextBlob
• Machine Learning Models
• Word Embeddings

About the Dataset:

• Source: Crowflower Data for Everyone
• Twitter data of February 2015 was scraped, the tweets are classified into positive, negative & neutral emotions

Insights from the Data

Sentiment Distribution Analysis

63% of the tweets are negative, whereas 21% positive & 16% neutral

Analysis of Airlines mentioned in the tweets

United Airlines followed by U.S Airlines have the highest tweet mentions

Analysis of Top 10 Tweet Locations

New York location has the highest tweet mentions

Analysis of User Time-zones

Eastern time followed by Central time are the timezones of most users

Word Cloud

Word Cloud of Positive Reviews

Word Cloud of Negative Reviews

Analysis of Negative Tweets

Top 15 Unigrams

Top 25 Bigrams

Top 25 Trigrams

Results of VADER Sentiment Analysis

Accuracy of VADER Lexicon is 49.61%

Results of TextBlob Analysis

Accuracy of TextBlob is 42.91%

Results of Machine Learning Models

Results of Word Embeddings (word2vec) Sentiment Analysis

Inference:

The following models have the best performance

• Sentiment Analysis with Logistic Regression gives 78% accuracy & 0.77 AUC-ROC Value
• Word Embeddings with Gradient or XG Boost are good classifiers with 77% accuracy & 0.73 AUC-ROC Score