|"Stance Detection" |
|Speaker ||: ||Dr. Kareem Darwish |
|Date ||: ||29 April 2019 (Monday) |
|Time ||: ||13:30 |
|Location ||: ||Faculty of Computer and Informatics Engineering, Student Lab. |Abstract:
The talk presents several supervised, semi-supervised, and unsupervised methods for detecting the stance of social media users towards controversial topics. Supervised learning can be performed using a variety of classifiers, such as linear and deep neural network classifiers, and a diverse set of features such as text-level features (e.g., words or hashtags), user-interaction features (e.g., user mentions and retweets), and profile-level features (e.g., name and location). Other supervised methods include: collective classification and projection into a lower dimensional space prior to classification. Label propagation is an effective semi-supervised method that propagates labels in a network based on follow or retweet relationships or based on the sharing of identical tweets. Recent work projects user onto a lower-dimensional space then uses clustering to perform unsupervised stance detection. The presentation will highlight the strengths and weaknesses of different approaches and suggest best practices for stance detection.
Dr. Kareem Darwish is a principal scientist at the Arabic Language Technologies at QCRI with interest in natural language processing (NLP), social computing, and information retrieval. He was as a researcher at the Cairo Microsoft Innovation Lab and the IBM Human Language Technologies group in Cairo. He also taught in the Electrical Engineering Department at the German University in Cairo and the Faculty of Computer and Informatics at Cairo University. He is currently developing a state-of-the-art Arabic NLP toolkit that includes POS tagging, named entity recognition, parsing, etc. In social computing, he is working on the automated detection of propaganda accounts on social media and on stance detection.