Cluster and Sentiment Analyses on Customer Reviews

• Extracted customer sentiment from reviews using text mining
• Applied cluster analysis to identify segments with differentiating needs and characteristics
• Provided insights regarding which services need the most improvement

Dataset and Resources Used

R Version: 1.4.1717
Packages:
Data: Raw Data

Data Preparation

1. The raw dataset includes 17 variables
2. A subset of 8 relevant variables was selected
   • Review (free text)
   • Overall Rating (numerical)
   • Food and Beverage Rating (numerical)
   • Cabin Service Rating (numerical)
   • Seat Comfort Rating (numerical)
   • Entretainment Rating (numerical)
   • Value for Money Rating (numerical)
   • Ground Service Rating (numerical)
3. Imputation of Missing Values

Initial Sentiment Analysis Using the Overall Rating

• Exploration of the proportion of positive and negative words for each rating (1-10)
• Three general-purpose lexicons were evaluated to against ratings: bing, ncr, and AFINN
• The bing lexicon, as it presented a significant correlation with the overall rating
• Exploration also indicated a prevalence of passenger dissatisfaction

Dimension Reduction with Factor Analysis

• Through factor analysis, we found that using two components would explain 74% of the six rated categories

Cluster Analyses

• To determine the optimal number of clusters multiple methods
• Results from Analyses suggest a three-cluster solution

Hierarchical Clustering

K-Means Clustering

Sentiment by Clusters

• The highest prevalence of positive reviews was in Cluster 3
• The highest prevalence of negative reviews was in Cluster 1

Clusters Profiling

Profile Segments by Needs

Cluster 1:

• Represent the most unsatisfied passengers overall
• Reviews suggest that negative ratings are largely driven by:
  • Poor customer service
  • Flight delays
  • Overall poor experience across pre-flight(PA2), in-flight(PA1), and post-flight categories(PA2)

Cluster 2:

• More likely to have negative pre-flight experiences, such as flight delays or cancellations
• In comparison to Cluster 2, they are slightly more satisfied with the in-flight experience

Cluster 3:

• Represent the most satisfied passengers overall
• Provided relatively positive text reviews and ratings across all measures.
• Reviews are largely positive and primarily attributed to the in-flight experience (PA1)

Profiling Across Factors

Profiling Across Individual Variables

Common Words on Reviews

Profile Segments by Characteristics

Cluster 1:

• Least likely to travel in first or business class
• The least likely to have experienced a flight delay/cancellation or layover

Cluster 2:

• The most likely to have experienced a flight delay, consistent with common words
• The most likely to have a connecting flight

Cluster 3:

• Higher percentage of business or first class passengers
• They appear less likely to travel for leisure purposes, compared to Cluster 1 and 2