American Airlines Twitter Sentiment Analysis

Synopsis

This is another mini project I carried out to not only display my data mining and visualization skills but to also show my ability to gather and manipulate data from multiple sources. Due to the limitations the Twitter API has set on tweet pulls and date range, I was able to extract 1340 tweets generated within the last 7 days; from the date of tweet pull. I performed this analysis using the R Programming Language.

The goal of this report is to identify issues that American Airlines (AA) customers experience. These issues will be identified using a sentiment scoring analysis on tweets from AA customers. The sentiment scoring analysis will highlight negative and positive customer sentiments. The results are displayed using a word cloud of frequently used words/phrases and scatter plots to show customer sentiments.

Excluding the neutral sentiments (neither negative or positive), negative AA customer sentiments accounted for ~ 63% of the tweets pulled for this analysis while the positive AA customer sentiments accounted for ~ 37%.

The most frequent negative sentiments resulted from flight delays and cancellations, baggae claim issues and customer service/call center issues. Let’s not discount the few good sentiments but having negative reports that far outweight the positive goes to say it is imperative that improvements be made to correct underlying issues to poor customer experience.

Recommendations

Based on this analysis and the assumption that the AA customer experiences retrieved from the tweets pulled in this report are to an extent representative of the sentiments AA customers feel, I recommend the implementation of the following:

• Improving overall customer care:

  • AA should assign more personnel who are knowledgable of current events to attend to and better assist with customer needs and concerns.

  • AA should investigate baggage claim issues to improve lost baggage concerns.

• Research shows customers are twice as likely to share a negative experience than a positive one. Therefore, AA should help customers share positive experience reviews easily. For example, whenever a flight lands on time, AA’s IT system can automatically send emails or texts to the phones of passengers with a link to twitter and other media platforms with pre-completed positive sentiment texts. This process can also be fostered by flight attendants who can direct passengers to share their positive sentiments upon on-time flight landing.

Note: Due to the limit on the number of tweet pulls imposed by the Twitter API, the tweet data in this analysis is not representative of all AA customer experiences.

Analysis Breakdown

Shown below are the steps I took to carry out this analysis:

• Created Twitter application, setup API keys and tokens and connected to Twitter timeline using the ‘twitteR’ package in R
• Pulled tweets using ‘#americanairlines’
• Manipulated and restructured tweet texts
• Tokenized tweet texts and prepped data for visualization
• Visualized tweets using the ‘wordcloud’ package in R
• Carried out sentiment analysis to separate tweets into positive and negative categories
• Lastly, I visualized sentiment scores and positive and negative tweet words/phrases

Tweet Texts Summary

Let’s visualize the words for the most frequently tweeted complaints and praises.

Note: The size of the text corresponds to the number of times that text appeared in the sample of tweets pulled. I.e., the larger the text, the higher the frequency.

The word cloud above doesn’t do a good job of interpreting what AA customers are really trying to say so let’s dive into a sentiment analysis to have a better understanding of the tweets.

Sentiment Analysis

The scatter plot below clearly shows there are more negative sentiments than there are positive.

Displayed below are word clouds of the most frequent positive and negative sentiments.

That’s a much better representation of customer sentiments than the first wordcloud visualization. Here we can see that some people actaully do appreciate AA’s service.

Tweet Preview

To give you an idea of what AA customers have actually tweeted, I have created a table of the top 5 positive tweets and the top 5 negative tweets. These tables are shown below.

Top 5 Positive Tweets

##                                                                                          text
## 1              london heathrow flagship lounge   relaxing gr staff good food   nice respite  
## 2        lessons  jetairways    competitors  compassion empathy   basic good customer service
## 3                       love   a aircraft free movies the pitch   seats dramatically improved
## 4           a beautiful plane  i  flying  work   continents i love people  finding solutions 
## 5 every time i fly i   pleasant experience grateful  helpfulness excellent customer service

Top 5 Negative Tweets

##                                                                                                          text
## 1              no worst airline dirty filthy dirty plane hour   runway  pilot didnt calculate  winds disaster
## 2  unhelpful  cancelled  flight refuses  accommodate  refunding cancelled leg  flight forcing lengthy reroute
## 3                                                      bad hrs delay    plane due  maintenance sick upset aa 
## 4                                          guys   fucking worst hour delay   denver   i   mechanical problems
## 5                another chaotic lgayul trip disorganized grumpy gate agents dumpy terminal tired plane thx

Note: From data cleaning and manipulation, tweets have lost some structuring but still contain the main information

Further Work

• To build a larger dataset of tweets for a better understanding of customer experience
• To analyze customer experience by geographical location
• To build a library for natural language processing. E.g: “stuck for 6 hours”, “flight delayed for 6 hours” all grouped to one sentiment, “tardy” etc. This would help improve the word cloud visualization and the accuracy of the sentiment scoring.

Contact

email: nky@utexas.edu