Abstract:
In face to face communications, people often rely on nonverbal cues such as body
language, facial expressions, gestures, physical proximity, and dress to communicate and
establish relationships. Recently computer mediated communication became a popular
way of interaction. Unfortunately nonverbal elements are normally absent in online
communications.
This thesis presents an affect recognition model that assesses the emotional states of
online users from textual messages. The study is based on the Social Information
Processing (SIP) theory argument that “when most nonverbal cues are unavailable, as is
the case in text-based computer mediated communication, users adapt their language,
style, and other cues to such purposes”. The focus is on emotion recognition from online
nonverbal textual symbols/patterns of vocalics (e.g. the use of capitals and use of
punctuation “!” and “!!s!” or “?” and “???”, length of response e.t.c), and those of
chronemics (e.g. time to respond to an email or to a discussion posting or a reply to a chat
message e.t.c) that are used in text.
The model uses Naïve Bayes classifier to recognize six basic emotions (anger, disgust,
fear, happiness, sadness and surprise). The training set was developed based on the results
of an online study named “Emotion Recognition from Nonverbal Symbols for Enhancing
Social presence in Online Environments”, that was carried out to determine the l evel of
use and the meaning of various nonverbal textual symbols used by students during their
online communications. Two sets of training data were prepared, a dictionary of words associated with different categories of mentioned emotions and messages labeled with the
six basic emotions collected from student’s online chats and posts. The messages were
annotated by three student raters independently, the level of agreement was measured
using Fleiss Kappa (k), and the reliability of agreement among the raters was moderate
with a k of (0.7).
Results of a user study comparing a chat system integrated with the affect recognition
model with a conventional chat system suggest that an online interface that conveys
emotional information helps online users to interact with each other more efficiently thus
providing an enhanced social presence.