Predicting User Stances from Target-Agnostic Information using Large Language Models