I. Introduction
Change requests are a significant portion of any large enterprise's IT operations. Common examples of change requests are shown in Table I. The requests are submitted as unstructured texts via IT ticketing systems to a service desk. The requests are then fulfilled by Site Reliability Engineers (SREs) upon approval. This is a labor-intensive process. To that effect, enterprises have built platforms to automate this process. Such platforms rely on pre-approved automated changes with a pre-established set of change actions. With enterprises migrating their application workload to a cloud-native deployment and adopting continuous integration and continuous delivery (CI/CD), the service catalogs become outdated very fast. In practice, we found that the coverage of change actions in these service catalogs is poor. Several supervised [1]–[4] and semi-supervised approaches [5] have been proposed to identify actions from unstructured text. However, they do not solve the catalog coverage problem since they need a discovered list of actions.