Towards detecting TV customer frustration using Hidden Markov Models

Customer engagement has long been a major field of study in the market for digital products because a better understanding of customer behaviour enables service providers to optimise their offerings and retain the customers who are at risk of leaving. Process mining has been widely used in this context, as it is important to mine the changes in customer behaviour throughout the lifespan to observe how customers navigate the product or service. The application of Hidden Markov Models (HMM) to BT TV set-top-box (STB) customer viewing data is the main topic of this paper. We use continuous state values derived from event sequences to construct derived features from the source data. We then employ a sliding window-based approach to create features and facilitate smoothing of the outcomes. In order to improve the Internet Protocol Television (IPTV) service, the goal of this study is to explore the methodologies currently used for evaluating customer engagement. We suggest a workable strategy for identifying TV STB customer frustration. Also, we have discovered some intriguing insights regarding the inferred Gaussian distributions of four previously selected features concerning customers' behaviour in TV navigation journeys over time using the hidden states produced by the HMM, with one of the hidden states providing indicators of customer frustration.