An Algorithm of Query Expansion for Chinese EMR Retrieval by Improving Expansion Term Weights and Retrieval Scores

Abstract

Query expansion (QE) has been widely used in electronic medical record (EMR) retrieval for assisted diagnosis and clinical research. However, existing QE algorithms haven't achieved satisfactory performance in Chinese EMR retrieval, and one noticeable problem is that the weights of expansion terms and retrieval scores have unreasonable factors for lack of the solid consideration of clinical needs. Here we propose an algorithm of QE for Chinese EMR retrieval by improving expansion term weights and retrieval scores. First, the weights of expansion terms are assigned with semantic similarities, category weights and co-occurrence frequencies between expansion terms and multiple query terms. Then the retrieval scores calculated by expansion terms are limited to reduce the query drift caused by high-frequency expansion terms. Experiment results show that our method gets a 33.3% increase in the precision at top 10, a 90.4% increase in the recall, and a 13.2% increase in MAP compared with four baselines. It proves that our improvement scheme can ensure the accuracy of expansion term weights and decrease the query drift caused by QE, which substantially improves the performance of Chinese EMR retrieval.

Keywords

• Electronic medical record
• query expansion
• word2Vec
• co-occurrence
• BM25