CoMeT | <%=title %>
Collaborative Management of Talks Hello! sign in or register
Bookmark Talks, Share with Friends, and We Recommend More!
Advanced Search
Home
Calendar
Series
Speaker
Groups
Connections
 
 
 
 
Talk Detail
Posted: ISP Admin on  Jan 31 01:22:21 PM
Title: Designing an Interactive Tool for Natural Language Processing on Clinical Text  
Speaker:
Gaurav Trivedi
Intelligent Systems Program
Sponsor: University of Pittsburgh  >  School of Computing and Information  >  Intelligent Systems Program
Series: ISP Colloquium Series
Date: Feb 03, 2017 1:00 PM - 1:30 PM
URL:
Location: 5317 Sennott Square
Groups Posted: Human-Centered Computing Information Science Intelligent Systems Program 
Detail:

Using Electronic health record (EHR) data to inform decision support and support data-driven quality measures is an ongoing challenge, often requiring the use of Natural Language Processing (NLP) techniques to extract computable representations from unstructured free text. Although significant advances in NLP systems have increased the scope and power of these approaches, use of these tools often requires resource-intensive active collaboration between clinical experts familiar with the problem domain, and NLP and computational experts capable of constructing the necessary models. The goal is to develop tools that close this gap by providing interactive features for review and revision of NLP models. I will present the design of a prototype tool combining novel text visualizations to help users interpret NLP results modeling Boolean variables extracted from clinical notes, revise models and understand changes between revisions. The interactive NLP loop begins when a user starts reviewing a list of documents. Users can review the text along with the features identified by the system to predict the variables. They can then revise the prediction values by giving feedback. Our evaluation suggests that physicians with limited or no machine learning experience can use our tools to enhance models based on limited training data, leading to substantial improvements in model performance after as little as 30 minutes of time per variable.

Interest Area: Computer & Information Science & Engineering, General Interests, Health Sciences
 
 
People Who Viewed This Talk, Also Viewed
 
 
Export
RSS Feed: RSS 2.0
ATOM Feed: Atom
iCalendar: iCal
Share: Bookmark and Share
 
Google Calendar:
 
 
 
 
CoMeT Blog
©2009-2019 CoMeT - Supported by Google Grant
School of Information Sciences, University of Pittsburgh, 135 North Bellefield Avenue, Pittsburgh, PA 15260
Real Time Web Analytics