Push for New Users and New Content: Overview of Content Generation and Personalization of Large Language Model in Cold Start Scenario

School of Mathematics and Physics, Xi’an Jiaotong-Liverpool University, Suzhou, China

^* This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Push system plays an important role in personalized services of various platforms, but often due to the lack of new user behavior data and interactive data of new content, resulting in the cold start problem, the traditional system cannot accurately push such new users. In recent years, the rapid development of large language model (LLM) shows its advantages in text rich and data sparse environments by using the transmission based paradigm, semantic understanding and pre training knowledge. This paper systematically summarizes the core methods of LLM in new user portrait reasoning, push copy generation, new content understanding and push scheme construction. The technical paths based on cue engineering, context learning, retrieval enhanced generation (RAG) and meta learning are analyzed respectively, and the advantages and disadvantages of pure LLM and hybrid architecture are compared. Finally, this paper summarizes the four challenges of hallucination control, reasoning delay, privacy ethics and effect evaluation, and looks forward to the future direction of the combination of LLM and push system, so as to provide reference for the construction of the next generation of recommendation system with better effect.