Feature Engineering and Characterization of Classifiers for Consumer Health Information Search

摘要

Health information search (HIS) is the process of seeking health awareness information on the Internet by health professionals and consumers. Identifying whether the retrieved text is relevant to consumer query and identifying whether it supports, opposes or is neutral to the claim made by the query are challenging tasks in HIS. In this paper, we present our methodology to address these two tasks using supervised learning approaches by performing feature engineering and characterization of classifiers. We have used seven variations including an ensembling approach and hierarchical boosting by incorporating statistical feature selection to different set of features and have determined the best solutions to the two tasks. We have evaluated our methods using CHIS@FIRE2016 data set. We have obtained accuracies of 82.4% for the first challenge using hierarchical boosting and 61.48% for the second using ensembling method. These results are promising when compared with those of other systems.