Scenario

We are developing a scenario for the system that we can use as a model for system development (see http://www.cs.columbia.edu/diglib/PERSIVAL/scenario). The scenario is based on questions a resident might ask during round or that a cardiologist might have in reviewing a case. We have selected three actual patients, sanitized their records, and developed several contexts under which different end users (residents or cardiologists) would have questions. We have carried out searches for these patients in the Medline database based on the questions and identified which of the retrieved articles are actually relevant. We have a testbed of full-text articles only in the domain against which we have also run searches, which result in 6 possible relevant articles. A different subset of these articles has been identified as relevant for each patient. Finally, the summarizer has been run on each article, with the generic system output currently produced. In addition, we are developing manual target output that specifies which pieces of the article are relevant given different patients under consideration. The initial mock-up was done with input from physicians on our team. We are currently developing evaluation techniques to explore new scenarios that naturally occur by audio-taping rounds where many questions are asked that are specific to the patient under care.