MCompassRAG: Topic Metadata as a Semantic Compass for Paragraph-Level Retrieval

In three linesMCompassRAG improves RAG systems by using topic-level metadata as a semantic compass for paragraph-level retrieval. The method enriches chunk representations with topic signals in the same embedding space and trains a lightweight retriever via LLM-teacher distillation. Across six benchmarks, it gains 8.24% in information efficiency with 5× lower latency than efficient RAG baselines.

Summary generated by Claude — human-verified