Improving LLMs for Recommendation with Out-Of-Vocabulary Tokens | Read Paper on Bytez