Issue № 29 | 2017 (3)
Original research data in MEDLINE database allows one to find the most comprehensive answer to clinical query. This entails the problem of large amounts of material that needs to be analyzed. The main study objective was to develop an algorithm for search results ranking of medical studies based on the levels of evidence. Developed search engine is based on a combination of classifiers that determine the level of evidence and subtype of medical intervention for a study abstract. As the main classification algorithms linear classifiers, as well as AdaBoost with Random Forest and SVM with RBF kernel were considered. Evaluation of the quality of classification by medical intervention subtypes was obtained using 5-folds cross validation method. Generative probabilistic LDA model was used for solving the problem of training set imbalance. The developed algorithm allows one with 92% precision to determine the study level of evidence, and to rank query results in relevance descending order.
evidence-based medicine, algorithm, MEDLINE, searching.
Kamalov M. V., Dobrynin V. Yu., Balykina Yu. E., Kolbin A. S., Verbitskaya E. V. Ranking Algorithm for Medical Research Results Based on the Levels of Evidence at the Stage of Getting Answers to Search Queries. Medical Technologies. Assessment and Choice. 2017; 3(29): 11–21.