In knowledge-intensive and multi-hop query tasks, graph-based retrieval-augmented generation (GraphRAG) is highly effective. However, many existing methods primarily rely on entity-based graphs and implicit semantic relevance propagation, leading to two main issues: (i) under-retrieval when user queries are abstract and semantically sparse at the entity level, and (ii) brittle multi-hop reasoning where noisy activations can derail entity-to-entity transitions and corrupt the inferred relation chain, resulting in unreliable conclusions.
To address these challenges, we propose FlowRAG, a semantic-aware retrieval framework that enhances both semantic recall and explicit reasoning. Specifically, FlowRAG constructs a quad-level heterogeneous graph over passages, summaries, sentences, and entities, where summary nodes act as a coarse semantic hub.
During retrieval, a dual-granularity activation module combines summary-query alignment with sentence-level matching to robustly activate relevant entities under paraphrase and abstraction. We then introduce a frequency-aware weighted flow module that routes relevance through entity-passage links weighted by within-passage term frequency, pruning noisy connections and extracting high-confidence reasoning paths as an explicit logic skeleton for generation. Extensive experiments demonstrate that FlowRAG achieves state-of-the-art performance on complex reasoning benchmarks.
Blogger's Review: FlowRAG effectively addresses the shortcomings of traditional GraphRAG methods by constructing multi-layer heterogeneous graphs and introducing a frequency-aware module. This innovative design not only enhances the model's reasoning capabilities but also provides a more reliable solution for knowledge-intensive tasks, showcasing significant application potential.