MMINN: Management-Model Informed Neural Network with Interpretability and Causal Relationship

Deep neural networks have achieved near-human accuracy levels in various classification and prediction tasks, such as in processing image, text, speech, and video data, while the scholars in the management research may have never widely recognized the method. The lack of use may be because the current neural network is still treated as the black box, which lacks the interpretability and the capability to explain the causal relationship in the observational social phenomenon. To extend this method to the management field, this paper puts forward an improved neural network model based on the management theories regarding decision-making strategies. Large-scale datasets of complex project management game simulations are made interpretable by introducing the improved deep learning methods, and causal relationships are explained by incorporating time series methods. This is based on the positivism in social science that the variables in management model should be with the time-series relationships. However, what the result of traditional empirical method based on statistics deduced is the correlation rather than the causality. To improve the methodology in management field, we take a project decision-making simulation game as the research object and conduct experiments using the data collected by Yu [13] on teamwork decision-making. This paper finally constructs a deep learning method based on decision-making strategy, creating a new research paradigm in management. Our method achieves a significant and consistent improvement as compared to other baselines.