Table 4 Comparison of classical chunk-selection baselines versus our DQN-based selector.

From: An integrating RAG-LLM and deep Q-network framework for intelligent fish control systems

Method

Precision@1

Recall@5

QA accuracy (%)

BM25

0.65

0.85

60.0

MMR (\(\lambda =0.7\))

0.68

0.88

63.0

Top-k Cosine (\(k=5\))

0.70

0.89

65.0

DQN

0.78

0.94

72.0

  1. Metrics reported are precision@1 (top-1 retrieval precision), recall@5 (correct chunk in top-5), and downstream QA accuracy when those chunks are passed to the LLM. Significant values are in bold.
Back to article page