Using discharge letters and context representation in information retrieval of medical literature

Electronic discharge letters are in many cases insufficient as a recipe for further treatment and recovery. Through using discharge letters as input to our neural network based information retrieval system Pasent, we have been able to provide relevant medical information. The Pasent search method uses predefined knowledge about the context, paired with the vocabulary of the input document, to compute a relevance measure for a potential result document. In the reported experiments, we achieved search results comparable to the tftdf method without building specialised context models for the experiment.