A Survey of Mixture Models for Scientific Literature

The emergent abilities of large language models (LLMs) in knowledge understanding and problem reasoning has garnered considerable interest from both academic and industrial sectors. The data processing of LLMs has expanded from textual data to multimodal data, including images, formulas, videos, and charts, leading to a sharp increase in model parameters and computational power requirements. Mixture models are one of the solutions to these issues, but research on mixture models for the understanding of scientific literature data remains scarce. This paper outlines the application of LLMs in the field of knowledge engineering and analyzes the importance of multi-model data LLMs in the intelligence domain. Then, this paper first conducts a review analysis of research on mixture models in scientific literature understanding, analyzing respective research directions and related work from three aspects: multimodal data understanding, dynamic expert scheduling strategies, and expert domain-specific training. Finally, this paper summarizes the current state of research on mixture models for scientific literature and looks forward to specific research directions in this field. Scientific literature, as an important carrier of human knowledge, is the core key supporting innovation. Existing research mainly focuses on the general application of mixture models. For the first time, this paper conducts a review analysis of mixture model research specifically for the field of scientific literature.