[KongchangAI]
Benchmark

NarrativeBench

A benchmark for evaluating the multilingual narrative balance capability of large language models, comprising multilingual news articles and real queries and covering multiple controversial topics and query languages.

Timeline (last 90 days)

Nov 10

The researchers propose NarrativeBench, a benchmark for systematically evaluating LLMs' ability to present and balance diverse narratives in multilingual settings.

Unverified50%
Nov 10

NarrativeBench contains 1,333 news articles covering 23 publication languages and 23 controversial topics, and builds 460 queries based on real Reddit queries.

Unverified50%
Nov 10

The study evaluates 14 LLMs across 24 query languages.

Unverified50%
Nov 10

NarrativeBench's queries are derived from real user discussions on Reddit.

Unverified50%

All Facts (4)

Source Articles