Using Text Analytics for Customer Feedback Analysis in the Airline World – Part I
Customer satisfaction lies at the heart of every airline and a wide range of methods exist to measure it in order to generate insight. Historically, the majority of methods relied on customer feedback obtained through customer surveys. In the era of social media however, an abundance of customer feedback is available online. Customers now publish their reviews or ratings of their experience on numerous platforms like, Twitter, Facebook, TripAdvisor or Skytrax.
The challenge many airlines are facing today is to automatically process the vast amount of textual information. Thus, gaining valuable insight by considering all reviews is not only tedious but rather impossible task for a single human being. At zeroG we have thus taken a shot at a text mining approach.
Latent Semantic Analysis
For our analysis we examined thousands of publicly available reviews. Besides the review text, the customer also gave a quantitative feedback by rating the experience in categories like overall experience value for money, comfort, friendliness of staff on a scale of 1-10 and whether he would recommend the airline or not. Here at zeroG we used a technique called “latent semantic analysis” (LSA) to analyze the diversity of issues that lead to a positive or negative recommendation in a visual manner.
Latent semantic analysis (LSA) is a method from natural language processing that seeks to reveal relationships between a given set of documents. The basic assumption behind this approach is that words with a similar meaning will occur in similar pieces of text. As an example since we are analyzing flight reviews, we expect words like e.g. “aircraft”,” airport“, ”flight” to occur in a high proportion of reviews. LSA puts these words in buckets called topics and allows us to quantify the relevance of these topics in each user review.
To visualize the diversity of recommendation one can then create a scatter plot where each point represents an individual user review and the distance between two points can be interpreted as how similar the underlying review between the points are. Additionally color coding reviews with the customer’s positive or negative recommendation reveals an interesting fragmentation of the user reviews suggesting that LSA successfully catches the topics that drive the customer’s recommendation. Performing this analysis for example between two different shows a clear variation in the clustering of the positive negative feedback.
At zeroG we use this technique to constantly monitor customer feedback and identify trending topics. Not just to better understand the customer’s voice, but also to route feedback to the right departments.