<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div class="gmail_quote">Reminder: this afternoon!<br><br><a href="http://www.cs.cmu.edu/%7Enlp-lunch/" target="_blank">CL+NLP Lunch</a><br><b>Tuesday, Nov. 15 @ noon in GHC 4405</b><br>Lunch will be provided.<br><br>
<b>Diane Litman</b><br>
Professor, Computer Science<br>Senior Scientist, Learning Research and Development Center<br>
University of Pittsburgh<br><br><b>Automatically Predicting Peer-Review Helpfulness</b><br><br>One path to improving the quality of student writing has involved the<br>use of a peer review process, often supported by web-based<br>
technology. The long-term goal of our research is to use Natural<br>Language Processing to address three core problems in peer-review of<br>writing: reviews are often stated in ineffective ways, reviews and<br>papers do not focus on important paper aspects, and authors do not<br>
have a process for organizing paper revisions. This talk will present<br>our research on automatically predicting the helpfulness of peer<br>reviews, one important task for improving the quality of feedback<br>received by students, as well as for helping students write better<br>
reviews.<br><br>We first examine whether standard product review analysis techniques<br>also apply to our new context of peer reviews. We also investigate<br>the utility of incorporating additional specialized features tailored<br>
to peer review. Our preliminary results show that structural features,<br>review unigrams and meta-data are useful in modeling the helpfulness<br>of both peer reviews and product reviews, while peer-review specific<br>auxiliary features can further improve helpfulness prediction.<br>
Finally, we investigate how different types of perceived helpfulness<br>might influence the utility of features for automatic prediction. Our<br>feature selection results show that certain low-level linguistic<br>features are more useful for predicting student perceived helpfulness,<br>
while high-level cognitive constructs are more effective in modeling<br>experts' perceived helpfulness.<br><br>This work is done in collaboration with Wenting Xiong, Christian<br>Schunn, and Kevin Ashley, University of Pittsburgh.<br>
<br>Bio: Diane Litman is Professor of Computer Science, Senior Scientist<br>with the Learning Research and Development Center, and faculty with<br>the Graduate Program in Intelligent Systems, all at the University of<br>
Pittsburgh. She has been working in the field of artificial<br>
intelligence since she received her Ph.D. degree in Computer Science<br>from the University of Rochester. Before joining Pitt, she was a<br>member of the Artificial Intelligence Principles Research Department,<br>AT&T Labs - Research (formerly Bell Laboratories). Dr. Litman's<br>
current research focuses on enhancing the effectiveness of educational<br>technology through the use of spoken and natural language processing,<br>affective computing, and machine learning and other statistical<br>methods. Dr. Litman has been Chair of the North American Chapter of<br>
the Association for Computational Linguistics, has co-authored<br>multiple papers winning best paper awards, and has been awarded Senior<br>Member status by the Association for the Advancement of Artificial<br>Intelligence.
</div><br>
</body></html>