Extracting Information about Medication Use from Veterinary Discussions

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1452–1458, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Extracting Information about Medication Use from Veterinary Discussions Haibo Ding School of Computing University of Utah Salt Lake City, UT 84112 hbding@cs.utah.edu Ellen Riloff School of Computing University of Utah Salt Lake City, UT 84112 riloff@cs.utah.edu Abstract Our research aims to extract information about medication use from veterinary discussion fo- rums. We introduce the task of categoriz- ing information about medication use to deter- mine whether a doctor has prescribed medica- tion, changed protocols, observed effects, or stopped use of a medication. First, we create a medication detector for informal veterinary texts and show that features derived from the Web can be very powerful. Second, we cre- ate classifiers to categorize each medication mention with respect to six categories. We demonstrate that this task benefits from a rich linguistic feature set, domain-specific seman- tic features produced by a weakly supervised semantic tagger, and balanced self-training. 1 Introduction Natural language processing holds great promise for automatically extracting empirical data about med- ications, from the perspective of both doctors and patients. A wealth of information about the admin- istration and effectiveness of medications lies within unstructured text, including medical records created by health care professionals (e.g., discharge sum- maries) as well as informal texts written by medical practitioners and patients (e.g., Web forums). Previous work has been done on detecting med- ication terms and recognizing relations between medications and other medical concepts, such as dis- eases and symptoms. Our research explores a new problem: identifying and categorizing contexts in- volving the administration of medications, which we call medication use categorization. For each men- tioned medication, we want to know whether it was used in a patient’s care, and if so, what actions or ob- servations are being reported. Our task aims to dis- tinguish between contexts where a doctor prescribed a medication, changed the protocol of a medication (e.g., dosage or frequency), stopped use of a medi- cation, observed effects produced by the medication, or is asking a question about a medication. Distin- guishing these contexts is an important step toward being able to extract empirical data about medica- tion use, such as effectiveness, success under differ- ent protocols, and adverse events. Our research studies veterinary discussion fo- rums, which often contain informal vocabulary such as shortened and abbreviated medication terms (e.g. “pred” instead of “prednisone”, or “abx” for “an- tibiotics”). The first part of our research addresses the problem of medication detection from informal text. We create an effective medication detector us- ing supervised learning with linguistic features as well as contextual features acquired from the Web. We show that the Web context features substantially improve recall, and yield an effective medication de- tector even with small amounts of training data. Second, we design supervised classifiers for med- ication use categorization. We incorporate a rich set of contextual, syntactic, and sentential …

Content truncated to 500 words