{ "id": "llm/rag-hybrid-search-score-mismatch", "signature": "RAG hybrid search returns no results because dense and sparse scores are on different scales", "signature_zh": "RAG 混合搜索返回空结果，因为稠密和稀疏分数不在同一尺度上", "regex": "RAG hybrid search returns no results.*dense and sparse scores.*different scales", "domain": "llm", "category": "data_error", "subcategory": null, "root_cause": "When using hybrid search (dense + sparse), the dense embedding similarity scores (e.g., cosine similarity 0.0-1.0) and sparse BM25/SPLADE scores (e.g., 0-20) are not normalized to the same range before fusion, causing one component to dominate or return no results after thresholding.", "root_cause_type": "generic", "root_cause_zh": "当使用混合搜索（稠密 + 稀疏）时，稠密嵌入相似度分数（例如，余弦相似度 0.0-1.0）和稀疏 BM25/SPLADE 分数（例如，0-20）在融合前未归一化到同一范围，导致一个组件占主导地位或在阈值处理后返回空结果。", "versions": [ { "version": "langchain==0.2.5", "introduced": null, "deprecated": null, "removed": null, "behavior_change": null, "status": "active" }, { "version": "chromadb==0.5.0", "introduced": null, "deprecated": null, "removed": null, "behavior_change": null, "status": "active" }, { "version": "qdrant-client==1.9.0", "introduced": null, "deprecated": null, "removed": null, "behavior_change": null, "status": "active" }, { "version": "elasticsearch==8.13.0", "introduced": null, "deprecated": null, "removed": null, "behavior_change": null, "status": "active" } ], "os_specific": {}, "dead_ends": [ { "action": "Lowering the similarity threshold to 0.0 to force results", "why_fails": "This returns all documents regardless of relevance, defeating the purpose of hybrid search and polluting the context with irrelevant documents.", "fail_rate": 0.8, "condition": "", "sources": [] }, { "action": "Switching to only dense search", "why_fails": "This eliminates the sparse component, which may be crucial for keyword-based retrieval in domains like legal or medical where specific terms matter.", "fail_rate": 0.65, "condition": "", "sources": [] }, { "action": "Increasing the alpha weight for the sparse component to 0.9", "why_fails": "This does not solve the scale mismatch; it only shifts dominance. The scores are still on different scales, so the fusion remains skewed.", "fail_rate": 0.7, "condition": "", "sources": [] } ], "workarounds": [ { "action": "Normalize both score sets to [0,1] using min-max scaling before fusion. Example:\n\ndef normalize_scores(scores):\n min_s, max_s = min(scores), max(scores)\n if max_s == min_s:\n return [0.0] * len(scores)\n return [(s - min_s) / (max_s - min_s) for s in scores]\n\ndense_scores = normalize_scores(dense_scores)\nsparse_scores = normalize_scores(sparse_scores)\nhybrid_scores = [alpha * d + (1-alpha) * s for d, s in zip(dense_scores, sparse_scores)]", "success_rate": 0.9, "how": "Normalize both score sets to [0,1] using min-max scaling before fusion. Example:\n\ndef normalize_scores(scores):\n min_s, max_s = min(scores), max(scores)\n if max_s == min_s:\n return [0.0] * len(scores)\n return [(s - min_s) / (max_s - min_s) for s in scores]\n\ndense_scores = normalize_scores(dense_scores)\nsparse_scores = normalize_scores(sparse_scores)\nhybrid_scores = [alpha * d + (1-alpha) * s for d, s in zip(dense_scores, sparse_scores)]", "condition": "", "sources": [] }, { "action": "Use a rank-based fusion (RRF) instead of score-based fusion. RRF combines ranks directly, avoiding scale issues entirely. Example:\n\ndef reciprocal_rank_fusion(dense_ranks, sparse_ranks, k=60):\n scores = {}\n for rank, doc_id in enumerate(dense_ranks):\n scores[doc_id] = 1 / (k + rank + 1)\n for rank, doc_id in enumerate(sparse_ranks):\n scores[doc_id] = scores.get(doc_id, 0) + 1 / (k + rank + 1)\n return sorted(scores.items(), key=lambda x: x[1], reverse=True)", "success_rate": 0.95, "how": "Use a rank-based fusion (RRF) instead of score-based fusion. RRF combines ranks directly, avoiding scale issues entirely. Example:\n\ndef reciprocal_rank_fusion(dense_ranks, sparse_ranks, k=60):\n scores = {}\n for rank, doc_id in enumerate(dense_ranks):\n scores[doc_id] = 1 / (k + rank + 1)\n for rank, doc_id in enumerate(sparse_ranks):\n scores[doc_id] = scores.get(doc_id, 0) + 1 / (k + rank + 1)\n return sorted(scores.items(), key=lambda x: x[1], reverse=True)", "condition": "", "sources": [] }, { "action": "Configure the vector database to use built-in score normalization. For Qdrant, set `quantization_config=models.ScalarQuantization(scalar=models.ScalarQuantizationConfig(type=models.ScalarType.INT8))` which normalizes scores internally.", "success_rate": 0.82, "how": "Configure the vector database to use built-in score normalization. For Qdrant, set `quantization_config=models.ScalarQuantization(scalar=models.ScalarQuantizationConfig(type=models.ScalarType.INT8))` which normalizes scores internally.", "condition": "", "sources": [] } ], "workarounds_zh": [ "在融合前使用最小-最大缩放将两组分数归一化到 [0,1]。示例：\n\ndef normalize_scores(scores):\n min_s, max_s = min(scores), max(scores)\n if max_s == min_s:\n return [0.0] * len(scores)\n return [(s - min_s) / (max_s - min_s) for s in scores]\n\ndense_scores = normalize_scores(dense_scores)\nsparse_scores = normalize_scores(sparse_scores)\nhybrid_scores = [alpha * d + (1-alpha) * s for d, s in zip(dense_scores, sparse_scores)]", "使用基于排名的融合（RRF）而不是基于分数的融合。RRF 直接合并排名，完全避免尺度问题。示例：\n\ndef reciprocal_rank_fusion(dense_ranks, sparse_ranks, k=60):\n scores = {}\n for rank, doc_id in enumerate(dense_ranks):\n scores[doc_id] = 1 / (k + rank + 1)\n for rank, doc_id in enumerate(sparse_ranks):\n scores[doc_id] = scores.get(doc_id, 0) + 1 / (k + rank + 1)\n return sorted(scores.items(), key=lambda x: x[1], reverse=True)", "配置向量数据库使用内置分数归一化。对于 Qdrant，设置 `quantization_config=models.ScalarQuantization(scalar=models.ScalarQuantizationConfig(type=models.ScalarType.INT8))`，这会在内部归一化分数。" ], "transition_graph": { "leads_to": [], "preceded_by": [], "frequently_confused_with": [] }, "official_doc_url": "https://qdrant.tech/documentation/concepts/hybrid-queries/#score-normalization", "official_doc_section": null, "error_code": null, "verification_tier": "ai_generated", "confidence": 0.87, "fix_success_rate": 0.8, "resolvable": "true", "first_seen": "2024-05-22", "last_confirmed": "2024-06-01", "last_updated": "2024-06-01", "evidence_count": 1, "tags": [], "locale": "en", "aliases": [] }