Machine Learning Helps Decipher Consumer Medical Searches

(USA.gov)

Many consumers turn to Web sites like WebMD for comprehensive health and medical information, but they cannot help as much if visitors searching the sites use unclear or unorthodox language to describe their conditions. A group of Georgia Tech researchers in Atlanta have created a machine-learning model that enables the sites to learn visitors’ dialect and other medical vernacular, to help the sites provide answers for visitors who search using such language.

Called dialect topic modeling (diaTM), the system learns by comparing multiple medical documents written in different levels of technical language. By comparing enough of these documents, diaTM eventually learns which medical conditions, symptoms and procedures are associated with certain dialectal words or phrases, thus shrinking the gap between consumers with health questions and the medical databases they turn to for answers.

To build diaTM’s capabilities in various types of medical terminology, the researchers pulled documents from WebMD and other online sources such as Yahoo! Answers, PubMed Central, the Centers for Disease Control and Prevention.

In small scale experiments, the researchers found that diaTM can achieve a 25 percent improvement in  normalized discounted cumulative gain (nDCG), a measure of the relevance of information retrieval in a Web search. In most studies of Internet search technology, say the researchers, a 5 percent improvement in nDCG is considered significant.

The results of the research are reported by Steven Crain, a Ph.D. student in computer science at Georgia Tech, who is the lead author of the paper “Dialect Topic Modeling for Improved Consumer Medical Search,” presented today (17 November) at the American Medical Informatics Association Annual Symposium in Washington, D.C. Collaboration and funding were provided by Oak Ridge National Laboratory, Microsoft, and Hewlett-Packard.

*     *     *