Machine Learning Points to Key Words in Online Hospital Reviews

Machine Learning Points to Key Words in Online Hospital Reviews

Lyle Ungar, professor in the Department of Computer and Information Science, has applied his expertise in machine learning and natural language processing to suss out the linguistic signatures of a variety of human experiences. From predicting emotional states from the language of Twitter users, to flagging posts on an addiction support group message board that indicate a user may be at risk of a relapse, Ungar’s work reveals insights that are hidden in plain sight, among the everyday words people use.

His most recent collaboration is with Penn Medicine researchers, who found that the word most associated with negative Yelp reviews of hospitals, was “told.”

They published their findings in the Journal of General Internal Medicine.

Among the one-star reviews the researchers saw that featured “told” were frustrations about information that was ostensibly shared (“They never told me the cost of any of the procedures”), anger at a lack of listening (“I told her I did not want to discuss it any more but she persisted to badger me.”) and feelings of futility (“Some idiot doctor examined me and told me there was nothing they could do for me.”).

“Oftentimes, words such as ‘told’ hint at a breakdown in communication,” said Anish Agarwal, MD, a National Clinician Scholars fellow and Emergency Medicine physician at Penn Medicine. “I suspect that patients are not feeling listened to or heard and this could be driving poor experiences and low reviews.”

Read more about the study at Penn Medicine News.