Toward Automatic Analysis of Conversations
| Event Date: | February 2, 2017 |
|---|---|
| Speaker: | Ming Chiu |
| Speaker Affiliation: | Charles R. Hicks Professor of Educational Psychology and Research Methodology, Purdue University |
| Type: | Research Seminar |
| Time: | 3:30 - 4:20 PM |
| Location: | Arms B071 |
| Priority: | No |
| School or Program: | Engineering Education |
| College Calendar: | Show |
Like running an obstacle course, automatic analysis of conversations requires overcoming many obstacles, including automation of transcription, categorization/coding and statistical analysis. After briefly discussing the first two issues, I detail my artificial intelligence expert system for statistical discourse analysis (SDA). SDA integrates multilevel, time-series analysis and statistical identification of pivotal moments and time periods within a conversation. To illustrate SDA, I test hypotheses on how sequences of group mates’ thinking (cognition) and executive thinking (social metacognition) affect a student’s cognition, using 1,330 asynchronous messages written by 17 students in an online forum linked to an education technology course for 13 weeks. Results showed that men gave more explanations but asked for fewer explanations. Also, opinions were often followed by elaborations, which were often followed by explanations. Beyond conversations, SDA can be used to identify online sexual predators, trace individual learning, detect corruption, and evaluate decisions during wars.
Bio | Ming Ming CHIU is Charles Hicks Professor of Educational Psychology and Research Methodology at Purdue University. A graduate of Columbia (BS, computer science), Harvard (EdM, interactive technology) and UC-Berkeley (PhD, mathematics education), he serves on the advisory board of mainland China’s Ministry of Education’s National Evaluation of Primary and Secondary Schools. He invented:
- statistical discourse analysis—called one of the best 50 learning science ideas by the International Society of the Learning Sciences—and
- multilevel diffusion analysis to detect corruption in the music industry.
He showed how economic growth, inequalities and cultural values affect nearly 500,000 students learning in 65 countries. Supported by 26 grants totaling over $4.5 million, he disseminated his research through 167 publications (92 journal articles), 3 television broadcasts, 17 radio broadcasts, and 148 news articles in 21 countries and regions. He is developing automatic analyses of online discussions and detection of online sexual predators.