Combining Query Reformulation and Re-ranking to Improve Query Expansion in Chinese EMR Retrieval

The methods of query expansion (QE) have achieved significant performance in information retrieval of electronic medical records (EMR). It is a pity that the direct addition of expansion terms may cause query drift, which decreases the precision of EMR retrieval. To solve the issue, we combine the methods of query reformulation and re-ranking to improve the performance of QE in Chinese EMR retrieval. First, the synonyms and hyponyms are extracted from a Chinese medical knowledge graph as expansion terms, and the weights of expansion terms are calculated with the combination of semantic similarities, category weights, and co-occurrence frequencies. Then the query is reformulated by these expansion terms, and the EMR documents are retrieved by the expanded query. Second, four categories of medical terms in top retrieval results are selected, and the performances of all combinations in re-ranking are tested to select the most proper terms for re-ranking. Then the original retrieval results are re-ranked by these terms. Experiments show that the algorithm can promote the effectiveness of EMR retrieval compared with baselines, which shows that the combination of query reformulation and re-ranking can differentiate expansion terms of different categories and maximize their effects.