MTRec: Learning to Align with User Preferences via Mental Reward Models | Read Paper on Bytez