From Embeddings to Explanations: Reviewing Semantic Modeling in LLM-Centered Interactive Systems

This review examines semantic modeling in large language model (LLM)centered interactive systems, focusing on embeddings, knowledge integration, and explanation generation. We summarize how LLMs produce contextual embeddings for semantic understanding, including applications in dense retrieval and label embedding. We then survey methods for integrating structured and multi-modal knowledge, with emphasis on knowledge graphs and retrieval-augmented generation (RAG). Finally, we analyze explanation techniques such as chain-of-thought prompting and self-rationalization in dialogue systems. Empirical studies in domains like finance and health informatics demonstrate how semantic modeling supports interpretability and factual grounding. All findings are drawn from published academic literature.