Detection of Customer Satisfaction on Unbalanced and Multi-Class Data Using Machine Learning Algorithms

With the increase of internet technologies, the opinions and suggestions of the previous customers are very important for both organizations and the people who supply them when purchasing goods and services. In particular, the spread of bad comments about goods and service providers on the internet is a harbinger of a great disaster for these companies. For this reason, companies want to compensate for their lack of customer feedback. They compete with each other to maximize customer satisfaction. However, as the number of feedbacks increases, their work becomes more difficult. In this study, it is aimed to determine customer satisfaction automatically in order to overcome this difficulty. The study was conducted on the text data of six different airline companies. These data consist of three different class labels: positive, negative and neutral. Natural Language Processing (NLP) steps, which we can call as text mining, are processed on these data. Using different Machine Learning (ML) algorithms, the results are examined with tables and graphs. The unbalanced distribution of data make difficult the success of the algorithms used in the study. It is estimated that this study is a reference source for literature that will attract the attention of researchers interested in natural language processing and machine learning algorithms, and importantly most many companies.