UCD IIRG at TREC 2012 Medical Track

This paper describes the participation of UCD IIRG in the TREC 2012 Medical Records track, which fosters research in the retrieval of electronic health records using free text fields. Our contributions to this track investigate several problem areas in the retrieval of medical documents. Multiple knowledge sources are investigated to alleviate the issue of vocabulary mismatch. Medical records are verbose documents that give a full picture of a patient’s medical status including their family health information and their own medical history. A Condition Attribution and Temporal Grounding system is implemented to address such occurrences. A rule-based system is employed in order to extract the patient’s demographic information from their medical record. All extracted information is then leveraged using Indri’s structured query language. These methods are combined to identify patients who fit the exact criteria as described in natural language queries.