Add dataset card with links to paper and code

README.md

@@ -0,0 +1,21 @@
+---
+task_categories:
+- question-answering
+- text-generation
+language: en
+tags:
+- commonsense-reasoning
+- benchmark
+---
+
+# SaliTrap: Revealing the Salience Bias of Large Language Models in Commonsense Reasoning
+
+This dataset contains the SaliTrap benchmark, introduced in the paper [Would You Walk to the Car Wash? Revealing the Salience Bias of Large Language Models in Commonsense Reasoning](https://huggingface.co/papers/2607.28478).
+
+- **Code**: [https://github.com/Wuzheng02/SaliTrap](https://github.com/Wuzheng02/SaliTrap)
+
+## Description
+
+SaliTrap is a high-quality benchmark designed to evaluate the **Salience Bias** of large language models in commonsense reasoning. The dataset spans four trap dimensions: missing prerequisite, environmental mismatch, temporal/physiological violation, and rule mismatch. It consists of 1,145 items and is intended to test whether models rely too heavily on explicit distractors while ignoring implicit knowledge.
+
+For more details, see the paper and the code repository.