Liu, Wanru
(2020)
Sentiment Analysis of Twitter Data to Explore Customers’ Feedback Towards US Airline Services.
[Dissertation (University of Nottingham only)]
Abstract
Among the competitive airline industries, understanding the customers' opinions and improving them accordingly is very important for improving customer satisfaction to retain old customers and attract new customers. This research aims to carry out sentiment analyses on Twitter data relevant to the US airlines to determine the drivers behind the sentiment of each customer review. The identification of the drivers allows appropriate recommendations to be suggested to each airline to improve their services and the utilisation of machine learning methods enables the generation of an accurate model that could predict and classify the sentiments of customer reviews.
The research process begins with downloading customer reviews related to tweets data from the Kaggle dataset, which includes 14,640 tweets relate to six major US Airlines. With the use of preliminary charts, according to the results, negative sentiment accounts for the majority of each airline’s tweet reviews. Hence, the utilisation of Word Cloud and other visualization tools had led to the discovery of the reasons for negative tweets and the findings of this research revealed negative sentiment drivers of the US airlines which include customer services, late flights, cancelled flights, lost or damaged luggage, and flight booking problems, etc. Then, based on these reasons, related recommendations can be provided to improve the customer satisfaction of each airline.
Then, to effectively monitor and classify customers' feedback, a model has been built in this project to achieve accurate prediction and efficient classification of customer reviews. At first, the 80-20 rule was used to split the data. After that, two text feature extraction methods (Bag of words and TF-IDF) were selected. The results generated from each of the text feature extraction methods were used separately in four classifiers (Support Vector Machine, LightGBM, Naïve Bayes, and Random Forest) to build models that predict tweet sentiments. In the end, each extractor-classifier combination was tested for accuracy and was compared against each other. From the comparison, the most accurate predicted model is the SVM classifier based on TF-IDF feature extraction.
In conclusion, performing sentiment analysis to customer reviews on Twitter of airlines can provide valuable insights to airline owners on understanding the factors that influence customer reviews and help the airline owners to make data-driven strategic decisions.
Keywords: Sentiment Analysis, Twitter, Customer Feedback, Data Mining, Machine Learning Methods
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)