Satria Fadil Persada

A Naïve Bayes Based Machine Learning Approach and Application Tools Comparison Based on Telephone Conversations

This paper investigates the application of hybrid Bayesian based semi-automated task analysis machine learning tool, Text Miner. The tool is still in its developing stage. Telephone’s dialog conversation between call center agent and customer was used as training and testing dataset to feed in parsing based Text Miner. Preliminary results extracted from Text Miner based on the naive Bayes approach was further compared to one open sourced machine learning program, tokenizing based Rapid Miner. Fifteen prediction words combinations were compared and the study finds that both tools are capable of processing large dataset with Text Miner performs better than Rapid Miner in predicting the relationship between prediction words and main subtask categories.