A Development of Life-like Indonesian Conversational Agent
Internal Publication JSPS “Japan-East Asia Network of Exchange for Students and Youths-JENESYS Program” Report July 2009
Host University: Tohoku University, Japan
Host Professor: Tetsuo Kinoshita
URL: http://www.ka.riec.tohoku.ac.jp/index-e.html
Abstract
This paper briefly describes current development of a
life-like Indonesian conversational agent using machine learning
techniques. The key toward believable life-like agent is incorporating
emotions into an intelligent agent. In order to do that, our
text-based conversational agents need to classify text input given
by human user and respond emotionally correct.
However using available tools such as NB, MaxEnt and SVM,
our experiments for emotion-related text classification showed
accuracies below 60% which suggest the need of deeper investigations
on how to increase this accuracy. Regarding incorporating
emotions into an agent, some researchers have proposed to extend
rational agent’s BDI architecture. The experimental results are
promising and encourage us to explore further.
Index Terms—Life-like NPC, Conversational Agent, Emotion,
BDI
