Does AI-based sentiment analysis predict consumer purchasing behavior in subscription services?

Abstract

This essay investigates the use of AI-written text sentiment analysis as a predictor of consumer churn in the digital streaming economy. The authors employed over 210,000 records of users of Netflix to generate the primary sentiment scores of the raw reviews left by the user by implementing a Natural Language Processing (NLP) algorithm. These were then inputted into a Logistic Regression model that predicted binary cancellation possibilities. Despite the fact that preliminary values of the Weight of Evidence (WOE) factor have allowed establishing a mathematical relationship between the presence of extremely negative mood and subscription termination, the predictive factor has a final Area Under the Curve (AUC) prediction of 0.4944, which is analogous to a shot in the dark. The extremely low Total Information Value (0.0014) demonstrates that the human behaviour in purchasing is highly multidimensional. It is discovered that, as much as sentiment analysis is applied in the establishment of the overall brand health, text sentiment cannot be applied mathematically to engineer a commercially viable churn prediction system without integrating with other behavioral metrics.