Evaluating Automatic Summaries of Meeting Recordings

摘要

The research below explores schemes forevaluating automatic summaries of businessmeetings, using the ICSI MeetingCorpus (Janin et al., 2003). Both automaticand subjective evaluations werecarried out, with a central interest beingwhether or not the two types of evaluationscorrelate with each other. Theevaluation metrics were used to compareand contrast differing approaches to automaticsummarization, the deteriorationof summary quality on ASR output versusmanual transcripts, and to determinewhether manual extracts are rated significantlyhigher than automatic extracts.