Existing methods make predictions by encoding attribute contents through similarity but lack dynamic modeling of fine-grained attribute scenarios, resulting in a deviation in user interest ...