Sentiment analysis of unstructured customer feedback for a retail bank

Abstract

With the explosive growth of the Internet and social media, the communication model between an organisation and its customers has become increasingly complex. A problem arises due to the sheer volume of unstructured data that has to be processed for the purposes of studying and addressing customer feedback. This calls for the development of automated methods. Important objectives of such methods include the detection of the underlying sentiment of customer feedback, as well as the synthesis and presentation of this sentiment in meaningful clusters such as topics and geographical locations. In this paper, a case study is conducted in which unstructured customer reviews related to products and services of a South African retail bank are evaluated by means of sentiment analysis. After suitable preprocessing techniques are applied to the reviews, the process of developing suitable models (primarily within the realm of machine learning) for detecting sentiment with a high level of performance is described. Subsequently, model results are analysed, synthesised and visualised in order to extract valuable insight from the data. The findings of the study show that custom learning-based models significantly outperform both pre-trained and commercial tools in sentiment classification. Furthermore, the analysis approach is shown to yield actionable information that may inform decision making.