ravimeduri76/concept-bridge-benchmark

Concept Bridge Benchmark是一个用于评估大型语言模型（LLM）创造性语义推理能力的基准数据集。该数据集包含80个谜题，分为8个难度级别，要求模型在给定的起始概念和目标概念之间构建一个多跳链，使用指定的语义关系。数据集的结构包括不同难度级别的谜题，每个级别有不同的跳跃次数、可用关系和约束条件。此外，数据集还提供了评估协议、基线结果、关系词汇表以及引用信息。

The Concept Bridge Benchmark is a benchmark for evaluating the creative semantic reasoning abilities of large language models (LLMs). The dataset contains 80 puzzles across 8 difficulty levels, requiring models to construct multi-hop chains between given start and target concepts using specified semantic relations. The dataset structure includes puzzles of varying difficulty levels, each with different numbers of hops, available relations, and constraints. Additionally, the dataset provides an evaluation protocol, baseline results, a relation vocabulary, and citation information.