Description variables et dashboard

This document presents an analytic study of airline reviews scraped on Skytrax's website. The aim is to understand which airlines receive travellers favors, and which criterias are most important, through the analysis of online reviews.

This study is divided in three tabs (click the for more info about specific elements)

• One tab dedicated to client satisfaction analysis, based on client's reviews of multiple criterias;

• One tab presenting a sentiment analysis of the reviews' content. For this we used Natural Language Processing (NLP), one of artificial intelligence's hot topic;

• One tab showing a geographical analysis of the reviews, by area / country, matching the airlines with the countries they are from.

Description variables et dashboard

The reviews were gathered by using a web scraper we built specifically for exploring the Skytrax website and extract reviews. In this process we managed to get130 000 reviews, on 493 airlines, from 202 countries, on a 17 years period (2002 to 2019).

The passengers give their review based on 8 criterias : overall, price / quality ratio, cabin service, entertainment, seat comfort, ground service, food / beverages quality. They can also leave a text review et indicate whether they would recommend the company or not.

Previous Next

Explication métrique

To visualize satisfaction in a simple and stable way we define a satisfaction score as follows :
On a scale from 1 to 10, grades 0 to 6 are the detractors. Grades 7 et 8 are considered neutral and grades 9 and 10 are the promoters.

We then compute the score as : \[ score = \frac{number \> of \> promoters - number \> of \> detractors}{number \> of \> answers} \]

Their are pros and cons to this method :

Advantages :

• Easy to measure, communicate, benchmark

• Stable : the extreme grades often present in this type of study do not impact the score too much

Cons :

• Do not account for demographic bias

• Mask grade distribution

This score measure is derived from the Net Promoter Score, mostly used in marketing, but the underlying question is different and we have a detailed grade by criteria, which makes the identification of action evers easier.

Note: Only the 60 most rated company are listed each time

Satisfaction by traveller type

Explication analyse sentiments

Below is a Wordcloud, whose idea is very simple : after removing "parasite" words from out text, like conjunctions and pronouns for example, as well as context words (like “flight” in our case), we display the words that are present the most, their size depending on their frequency.

The above figures initiate a sentiment analysis for the airlines for which we had the most reviews. This analysis word/feeling was done with the help of the NRC emotion lexicon, a reference in the field.

The first graph gives the breakdown of sentiments within each airline 's reviews, and the second one associates each of this feeling with a positive / negative aspect and underlines the opinions about each company.

In scientific literature, their are a number of studies about using NLP on user reviews, like :

• An article about airline reviews analysis, handling the topic of clustering from a mathematical point of view
• A sentiment analysis based on Twitter data
• A blog article on the same topic

Wordcloud

Description de la carte