1. Submission No. 17 An Affective Agent (AA) for Predicting Composite Emotions AAMAS Demo 2015 Submission No. 17 Demo site: http://www.linjun.net.cn/affectiveagent/ http://www.linjun.net.cn/affectiveagent/

2. Submission No. 17 Introduction to Affective Agent (AA) AA Game is a simple interactive agent game for predicting your composite emotions after reading a social news (currently we use the data from sina.com.cn and it can be extended to other news source. AA uses an affective agent to predict human beings' complex emotion compositions when they are reading online articles is an interesting experience to users, especially for e-commerce or entertainment website. Based on OCC emotional theory, AA uses explicit emotion appraisals and a historical group emotion dataset to predict a user's hidden emotion compositions. The historical group emotion data are based on web users' self-reported emotion labels of their feelings when reading news articles on sina.com.cn between 1 January to 30 June, 2013. The experiment results show that the artificially generated composite emotions of the agent are highly similar to real users' composite emotions.

3. Submission No. 17 Our AA Architecture The AA Architecture consists of four major components, including: 1) the Knowledge Base (KB), which stores historical emotional data; 2) the Sensory Processor (SP), which accepts the input of explicit emotion appraisals reported by users; 3) the Analytics Unit (AU), which executes the e- motion predicting logic; 4) the Emotional Space (ES), which presents the predicted composite emotion to user.

4. Submission No. 17 Random 100 news’ prediction result from AA Download: original data fileoriginal data file

5. Submission No. 17 How to play the AA game Step 1. read the following social news. Step 2. label your emotion appraisal for the news Player can label two types of main emotion (from Touched, Shocked, Curious, Angry, Funny, and Sad) and their strength (from 1 to 100). Then player will see our Agent's predicting result for his/her hidden emotions Step 3. and then player can rate a score for the results comparing to his/her real emotion composition.

6. Submission No. 17 Conclusion & Contribution Predicting user's composite emotions are important to designing believable artificial companions. In this demonstration, we showcase an Affective Agent, including its architecture and an online test-bed platform for this purpose. Calculated based on real emotion voting data collected from Sina society news website shows our AA can achieve high prediction accuracy. Our platform will be open for public interaction during the AAMAS demonstration session and stay online after the conference. Players’ interaction with AA can potentially replace the original sina site for collecting more accuracy emotion data.