Michiel van der Meer
An Empirical Analysis of Diversity in Argument Summarization
My EACL ‘24 paper on whether NLP models, including LLMs, are able to represent diversity among opinions when they are tasked with summarizing them. Our experiments show that these models have trouble with (1) dealing with data from various sources, (2) representing arguments shared by few people, and (3) aligning with subjectivity in human-provided annotations.