A minority opinion is a less common viewpoint in a set of texts, reviews, or comments. In opinion summarization, it is the dissent that can be lost when a model compresses many inputs into one fluent answer and overweights the dominant sentiment. The result may be technically “accurate” at a high level but misleading by omission, because edge-case complaints, rare risks, or important exceptions disappear.
In cyber security, this matters whenever untrusted user text is fed into an LLM or summarizer. Attackers can flood a corpus with biased content, or hide instructions inside text to steer what the model emphasizes. Defenders reduce this risk with provenance, topic-based summaries, diversity checks, and links back to original evidence. Preserving minority opinion helps stop a summary from becoming a trust-shaping filter that hides warnings users need to see.