Predicting and interpreting digital platform survival: An interpretable machine learning approach

Despite the substantial economic impact of digital platforms, research on platform risk evaluation has been sparse. In this study, we investigate whether online content can serve as leading indicators of digital platform survival. We employ machine learning techniques to extract features from three types of online content, that is, user generated content, platform generated content, and third party generated content and examine their utilities in predicting platform survival. Using a predictive XGBoost algorithm and data crawled from a leading web portals of digital platforms for online lending in China, we find online content are strong predictors of platform survival. Furthermore, we use casual forest models to reveal the differences among the three type of online content in terms of predictive utility. Interestingly, we find the presence of third-party generated content indicates lower probability of platform survival while the platform with more user generated content has higher chance to survive. The relationship between platform generated contents and platform failure is not significant. Based on the results, we provide practical implications for market managers and platform owners.