FIB Dataset

Factual Inconsistency Benchmark (FIB) is a benchmark that focuses on the task of summarization. Specifically, the benchmark involves comparing the scores an LLM assigns to a factually consistent versus a factual inconsistent summary for an input news article. For factually consistent summaries, human-written reference summaries are used to manually verify as factually consistent.

事实一致性基准（Factual Inconsistency Benchmark，简称 FIB）是一项专注于文本摘要任务的基准测试。该基准主要涉及比较大语言模型（LLM）对于输入新闻文章的事实一致性与事实不一致性摘要所赋予的分数。对于事实一致的摘要，采用人工编写的参考摘要进行手动核实，以确保其事实一致性。