关键词: 可理解性, 二次学习, 规则归纳, 数据审计

Abstract:

In recent years, machine learning techniques have been used to analyze psychology data, with the hope of finding comprehensible patterns to depict and adjust psychological behaviors of human beings. This paper proposes twice-learning style rule generation algorithms that combine the comprehensibility of rule-based learning methods and generalization performance of state-of-the-art learning methods like ensembles and support vector machines (SVMs). Experimental results show that outputs produced by twice-learning style algorithms are always much more accurate and in many cases are more comprehensible than those produced by traditional rule induction algorithms.

Key words: comprehensibility, twice learning, rule induction, data editing