Generating Temporal Relation Candidates Based on Open-domain Relation in Clinical Notes

Chae-Gyun Lim, Ho-Jin Choi

Abstract: From the point of view in natural language understanding, capturing the contextual information contained in a document is heavily influenced by identifying the relations between entities. Also, in the medical field, there are attempts not only to grasp the information of relations in the clinical notes but also to find out temporal relations among the medical events or observations. In this paper, we propose a method for generating the candidates which become temporal relations based on the open-domain information for extracting relations from medical documents. Open-domain relations can be extracted without specifying any type of relations in the given document. If a relation instance has either a time expression or time-related event, it can also be used to create candidates that have the potential to become temporal relations. We expect that the proposed method helps to better understand the contextual information about temporal relations in medical documents.