Identifying Latent Similarities among Near-Miss Incident Records Using a Text-Mining Method and a Scenario-Based Approach

This research focuses on supporting an analyst's activity of interpreting the contents of existing incident reports. During this activity, analysts are always predicting expected scenarios of the incidents at hand in comparing that with the actual development of the incidents reported therein. In order to learn lessons from a particular prior experience, analysts should be aware of the latent similarities among the incidents and should experience a breakdown called "expectation-failure" to let that incident be surely printed in their memory. To let the human analysts experience this breakdown, our system introduces a theory of Memory Organization Packets (MOPs) as a framework for explaining the dynamic memory structure of the human. By utilizing this idea as a basis for scenario-based expectation of human analysts and by integrating this idea with a text-mining method, a system for supporting an incident analysis is developed for a domain of medical incidents. Results of the experiments using our proposing system are presented, where the subjects are nurses working for a hospital. Based on those results, effectiveness of the system is discussed from various viewpoints by investigating into the protocols gathered from the subjects of the experiments.