Paper Reading # 17

Comments
Comment 1
Comment 2


Reference
Estimating User's Engagement from Eye-gaze Behaviors in Human-Agent Conversations
IUI'10 February 7-10,2010 Hong Kong, China

Summary
This paper talks about different strategies to see if a person is engaged in face-to-face conversations, and how people interact with each other through the use of technology. The main goal of the system described in this paper is to determine if the user is fully engaged in a conversation. The most complicated issue to resolve in their research is to make sure that the system must perceive nonverbal behavior such as facial gestures, body movements, and other key movements that could infer a person is not fully engaged in a conversation. According to their research there are many eye-tracking systems that are very stable, and therefore can be used in their complex system. The system or the agent-as the authors refer to it as- must be capable of reengaging a person if the person seems to have lost the full engagement in a conversation. The agent must do this by changing the topic when some eye-gazes infer the user is bored during a dialogue. In order to analyze their system the authors describe a Wizard-of-Oz approach to collect data. The scenario of the experiment is a salesperson in a mobile phone store. The experiment is set up such that there is a person called the user located in a room looking a screen where the salesperson is being projected. Outside of the room, there is another person called the observer. The observer can look at the user through a one way window. Both of these individuals will have a button to press. The instructions are different for both individuals. The user must press the button when he believes the salesperson description to be boring. The observer will press the button when he believes the user is bored by the salesperson's description. The analysis used in this experiment is called 3-grams. The results observed by the authors show that selecting the agent's behaviors according to engagement estimation turned out to be effective in this agent-human interaction.
Discussion
This paper is really interesting because it could be implemented in academia. It would be extremely beneficial for a teacher to know when their audience is not being engaged by his/her lectures. The Wizard-of-Oz approach in this interaction showed great results in the agent-human interaction. This type of research can have great benefits for people who are always presenting, and can provide feedback to improve their presentation skills.