Friday, October 12, 2012: 7:40 PM
Hall 4E/F (WSCC)
Research done on temporal expressions in clinical text has not been explored much. Results generated by analyzing temporal expressions would provide both patients and doctors with information in an organized manner. Thus, both of them could see any progress experienced, and in addition doctors could find patterns over a timeline to support clinical decision making. We have started to identify temporal expressions in a set of electronic clinical texts with the ultimate goal of designing an interactive system that patients can access to look at a timeline of their medical history. We have applied the following natural language processing techniques. First, clinical texts are tokenized and normalized by converting all the tokens to lowercase. Second, speech tagging is done using dictionaries to categorize tokens into nouns, prepositions and numbers. Numbers are a special case since they need to be further analyzed in context so as to tell whether or not they are indicators of time. Third, the tagged text is parsed to extract noun phrases and prepositional phrases. Our focus is on identifying candidates as potential temporal expressions such as “for 2 weeks”, “at noon”, and so on. The file of candidate temporal expressions generated by our system is then evaluated using precision and recall measures to determine how good our results are. We use this information to further improve our system. We have observed that identification of temporal expressions relevant to a patient's medical history is not trivial. However, it is feasible with further research and work.