The rapid advancement of Large Language Models (LLMs) has significantly enhanced text generation quality in Natural Language Processing (NLP). However, practical applications impose complex requirements, particularly in fields such as generating analysis reports. This study systematically reviews evaluation methods for LLMs and proposes a general framework for complex text generation, encompassing Generation Content, Prompt Dimension, Retrieval-Augmented Generation (RAG), and LLM Fine-tuning. We first introduce the Complex Text Generation Task Evaluation Paradigm. Based on this paradigm, we identify 15 sub-indicators with corresponding evaluation methods to comprehensively assess and improve LLM performance. Our research fills gaps in existing evaluation systems and provides a scalable framework for future studies, enhancing the applicability and impact of LLMs across various domains.