Identification of Risk Control Systems using anchor-words based Correlation Explanation (CorEx) model

One of the prominent accident models used to assess risk and hazards in various work systems is the chain of events model, which models the escalation of a hazard into an accident in the form of a chain of sequential activities. To break hazard propagation along the chain, various risk control systems (RCS) need to be developed and implemented at various nodes of the chain. Unfortunately, identification of the RCSs is a tedious and time-consuming process involving manual efforts of subject-matter experts. This paper is an attempt made to use data driven models to identify the RCSs. Using a semi-supervised text mining model called Correlation Explanation (CorEx), we propose a framework that can automate this process of identifying RCSs from incident investigation reports using sets of anchor words. This model shows high applicability of generating multiple hierarchical themes (RCS) from text data without the need of training datasets, thus achieving better interpretability compared to unsupervised topic models while achieving savings in time and effort.