Table 2.

Evaluating contextual similarity (higher is better) and information entropy (lower is better) of summaries generated from four different summarization approaches. The reported scores are mean and standard deviation calculated across 184 patient conversations in the Distress Analysis Interview Corpus-Wizard of Oz data set.

Methods	Contextual similarity, mean (SD)	JSDa, mean (SD)
Extractive summarization	0.672 (0.076)	0.387 (0.107)
Abstractive summarization	0.676 (0.076)	0.373 (0.102)
Abstractive over extractive summarization	0.669 (0.092)	0.385 (0.103)
Knowledge-infused abstractive summarization	0.689 (0.088)	0.360 (0.104)

^aJSD: Jensen Shannon Divergence.