HuaSLIM: Human Attention Motivated Shortcut Learning Identification and Mitigation for Large Language models F i n d i n g s - A C L