San Francisco State University at TREC 2014: Clinical Decision Support Track and Microblog Track
Abstract
The Clinical Decision Support Track in TREC 2014 involved identifying biomedical articles that could assist in answering generic clinical questions. This paper discusses the methodologies adopted by the system Runsystem2, that we built for answering these medical questions. Runsystem2 operates by translating a narrative of a patient's case report into a list of structured medical concepts which are then used to generate the query. The articles retrieved for the query are then reranked based on their ability to answer the three types of clinical questions studied in this track: diagnosis treatment and test. The experimental results demonstrate that the developed system performed close to the median performance on most metrics. Our approach to the ad-hoc retrieval task focused on query expansion, language and re-tweet filtering, and URL boosting. Query expansion used pseudo relevance feedback based on frequency of reoccurring terms. The various filters were then applied over the results from the expanded query after which the remaining tweets were re-ranked via URL boosting. The experimental results demonstrate that the best search effectiveness is obtained when all three techniques are employed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2014
- Accession Number
- ADA618585
Entities
People
- Aayush Bhandari
- Anagha Kulkarni
- James Klinkhammer
Organizations
- San Francisco State University