{ "datasetId": "ai-search-rag-finetuning-citation-sample-2026-07", "schemaVersion": "1.0.0", "createdAt": "2026-07-20", "language": "zh-CN", "license": "https://creativecommons.org/licenses/by/4.0/", "licenseScope": "DragonAI's observation design, run-level counts, URL normalization, hostname aggregation, overlap calculation, field definitions, and Chinese descriptions. Extracted citation strings are included only as observation records. Third-party pages, complete model answers, provider names, trademarks, and API payloads are excluded and remain subject to their respective terms.", "disclosureZh": "DragonAI 运营 AI 产品经理课程，并监测本域名在 AI 搜索回答中的可见性。本数据集包含目标引用为零的结果，不以自有域名命中为纳入条件。", "researchQuestionZh": "千问、Kimi、Gemini 回答同一个 RAG 与微调选型问题时，在本次点时样本中引用了哪些来源域名？", "query": "RAG和模型微调应该怎么选择？请给出决策条件、验证实验和可靠来源。", "observationWindow": { "startedAt": "2026-07-20T23:33:18Z", "endedAt": "2026-07-20T23:35:22Z" }, "target": { "domain": "course.dragonai.tech", "path": "/comparisons/rag-vs-fine-tuning", "publicationPhase": "prepublication" }, "methodology": { "providers": [ "qwen_dashscope_web_search", "kimi_builtin_web_search", "gemini_google_search_grounding" ], "repetitionsPerProvider": 3, "samePromptAcrossRuns": true, "nativeWebSearchRequired": true, "completedRunInclusionRuleZh": "仅纳入 API 成功返回、执行原生联网搜索并能提取展示层引用 URL 的运行。", "rawCitationStringDefinitionZh": "监测器从展示层提取的一条 URL 形字符串记为一条原始引用字符串。", "normalizationRuleZh": "移除直接附在 URL 后的‘)：说明文字’或‘):说明文字’及末尾标点，再在单次运行内按完整 URL 去重；不合并不同路径、查询参数或重定向目标。", "normalizedUrlOccurrenceDefinitionZh": "一次运行中的一个规范化 URL 记为一次 URL 出现；相同 URL 在不同运行再次出现会再次计数。", "hostnameDefinitionZh": "解析 URL 主机名并转为小写；不同子域名分别计数。", "targetAttributionRuleZh": "只有可解析到 course.dragonai.tech 目标域名的引用才计为目标引用；无法解析的重定向不计入。", "brandMentionRuleZh": "回答正文出现预设品牌实体 DragonAI、烛龙智元或目标域名时记为品牌提及。", "rawAnswerPublication": "excluded_except_extracted_citation_strings", "publicRunRecords": "included_without_answer_text" }, "aggregate": { "completedRuns": 9, "citationSurfaceRuns": 9, "attributableCitationRuns": 9, "rawExtractedCitationStrings": 77, "normalizedUrlOccurrences": 76, "distinctNormalizedUrls": 46, "distinctHostnames": 31, "targetCitationRuns": 0, "brandMentionRuns": 0, "hostnameSharedByAllThreeProviders": [] }, "providerResults": [ { "provider": "qwen_dashscope_web_search", "model": "qwen-plus", "completedRuns": 3, "rawCitationStringCountsByRun": [ 10, 10, 10 ], "normalizedUrlCountsByRun": [ 10, 10, 10 ], "normalizedUrlOccurrences": 30, "distinctNormalizedUrls": 10, "distinctHostnames": 6, "sameCitationUrlSetAcrossRuns": true, "targetCitationRuns": 0, "brandMentionRuns": 0, "hostnameOccurrences": [ { "hostname": "cloud.baidu.com", "citationOccurrences": 15 }, { "hostname": "lubuxun.blog.csdn.net", "citationOccurrences": 3 }, { "hostname": "post.smzdm.com", "citationOccurrences": 3 }, { "hostname": "qianfanmarket.baidu.com", "citationOccurrences": 3 }, { "hostname": "www.163.com", "citationOccurrences": 3 }, { "hostname": "zhuanlan.zhihu.com", "citationOccurrences": 3 } ] }, { "provider": "kimi_builtin_web_search", "model": "moonshot-v1-128k", "completedRuns": 3, "rawCitationStringCountsByRun": [ 3, 2, 6 ], "normalizedUrlCountsByRun": [ 3, 2, 5 ], "normalizedUrlOccurrences": 10, "distinctNormalizedUrls": 7, "distinctHostnames": 5, "sameCitationUrlSetAcrossRuns": false, "targetCitationRuns": 0, "brandMentionRuns": 0, "hostnameOccurrences": [ { "hostname": "arxiv.org", "citationOccurrences": 4 }, { "hostname": "blog.csdn.net", "citationOccurrences": 3 }, { "hostname": "zhuanlan.zhihu.com", "citationOccurrences": 1 }, { "hostname": "www.ud.hk", "citationOccurrences": 1 }, { "hostname": "www.zhihu.com", "citationOccurrences": 1 } ] }, { "provider": "gemini_google_search_grounding", "model": "gemini-2.5-flash", "completedRuns": 3, "rawCitationStringCountsByRun": [ 16, 10, 10 ], "normalizedUrlCountsByRun": [ 16, 10, 10 ], "normalizedUrlOccurrences": 36, "distinctNormalizedUrls": 29, "distinctHostnames": 22, "sameCitationUrlSetAcrossRuns": false, "targetCitationRuns": 0, "brandMentionRuns": 0, "hostnameOccurrences": [ { "hostname": "www.databricks.com", "citationOccurrences": 4 }, { "hostname": "medium.com", "citationOccurrences": 4 }, { "hostname": "arxiv.org", "citationOccurrences": 3 }, { "hostname": "cohere.com", "citationOccurrences": 2 }, { "hostname": "dev.to", "citationOccurrences": 2 }, { "hostname": "docs.aws.amazon.com", "citationOccurrences": 2 }, { "hostname": "galileo.ai", "citationOccurrences": 2 }, { "hostname": "www.actian.com", "citationOccurrences": 2 }, { "hostname": "cloud.google.com", "citationOccurrences": 2 }, { "hostname": "unstructured.io", "citationOccurrences": 1 }, { "hostname": "www.coreweave.com", "citationOccurrences": 1 }, { "hostname": "www.redhat.com", "citationOccurrences": 1 }, { "hostname": "docs.nvidia.com", "citationOccurrences": 1 }, { "hostname": "openreview.net", "citationOccurrences": 1 }, { "hostname": "www.braintrust.dev", "citationOccurrences": 1 }, { "hostname": "www.evidentlyai.com", "citationOccurrences": 1 }, { "hostname": "www.meilisearch.com", "citationOccurrences": 1 }, { "hostname": "www.oracle.com", "citationOccurrences": 1 }, { "hostname": "www.patronus.ai", "citationOccurrences": 1 }, { "hostname": "community.ibm.com", "citationOccurrences": 1 }, { "hostname": "learn.microsoft.com", "citationOccurrences": 1 }, { "hostname": "redis.io", "citationOccurrences": 1 } ] } ], "runs": [ { "provider": "qwen_dashscope_web_search", "model": "qwen-plus", "runIndex": 1, "observedAt": "2026-07-20T23:33:18Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 10, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://cloud.baidu.com/article/3363752", "https://cloud.baidu.com/article/3367936", "https://cloud.baidu.com/article/3411968", "https://cloud.baidu.com/article/4394886", "https://cloud.baidu.com/article/5543913", "https://lubuxun.blog.csdn.net/article/details/154753402", "https://post.smzdm.com/p/arz386lx", "https://qianfanmarket.baidu.com/article/detail/1195249", "https://www.163.com/dy/article/L296MIS50556LPY6.html", "https://zhuanlan.zhihu.com/p/1981394187950786445" ], "targetCited": false, "brandMentioned": false }, { "provider": "qwen_dashscope_web_search", "model": "qwen-plus", "runIndex": 2, "observedAt": "2026-07-20T23:33:18Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 10, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://cloud.baidu.com/article/3363752", "https://cloud.baidu.com/article/3367936", "https://cloud.baidu.com/article/3411968", "https://cloud.baidu.com/article/4394886", "https://cloud.baidu.com/article/5543913", "https://lubuxun.blog.csdn.net/article/details/154753402", "https://post.smzdm.com/p/arz386lx", "https://qianfanmarket.baidu.com/article/detail/1195249", "https://www.163.com/dy/article/L296MIS50556LPY6.html", "https://zhuanlan.zhihu.com/p/1981394187950786445" ], "targetCited": false, "brandMentioned": false }, { "provider": "qwen_dashscope_web_search", "model": "qwen-plus", "runIndex": 3, "observedAt": "2026-07-20T23:33:18Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 10, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://cloud.baidu.com/article/3363752", "https://cloud.baidu.com/article/3367936", "https://cloud.baidu.com/article/3411968", "https://cloud.baidu.com/article/4394886", "https://cloud.baidu.com/article/5543913", "https://lubuxun.blog.csdn.net/article/details/154753402", "https://post.smzdm.com/p/arz386lx", "https://qianfanmarket.baidu.com/article/detail/1195249", "https://www.163.com/dy/article/L296MIS50556LPY6.html", "https://zhuanlan.zhihu.com/p/1981394187950786445" ], "targetCited": false, "brandMentioned": false }, { "provider": "kimi_builtin_web_search", "model": "moonshot-v1-128k", "runIndex": 1, "observedAt": "2026-07-20T23:34:41Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 3, "rawCitationStringsSameAsCitationUrls": false, "rawCitationStrings": [ "https://arxiv.org/pdf/2502.04095)：提供了RAG与微调的比较分析", "https://blog.csdn.net/2401_85325726/article/details/154349326)：讨论了RAG与微调的核心区别和适用场景", "https://zhuanlan.zhihu.com/p/1940538557031781213)：提供了RAG和微调的场景分析和选择建议" ], "citationUrls": [ "https://arxiv.org/pdf/2502.04095", "https://blog.csdn.net/2401_85325726/article/details/154349326", "https://zhuanlan.zhihu.com/p/1940538557031781213" ], "targetCited": false, "brandMentioned": false }, { "provider": "kimi_builtin_web_search", "model": "moonshot-v1-128k", "runIndex": 2, "observedAt": "2026-07-20T23:34:41Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 2, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://arxiv.org/pdf/2502.04095", "https://blog.csdn.net/2401_85325726/article/details/154349326" ], "targetCited": false, "brandMentioned": false }, { "provider": "kimi_builtin_web_search", "model": "moonshot-v1-128k", "runIndex": 3, "observedAt": "2026-07-20T23:34:41Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 6, "rawCitationStringsSameAsCitationUrls": false, "rawCitationStrings": [ "https://arxiv.org/pdf/2410.17783", "https://arxiv.org/pdf/2502.04095", "https://arxiv.org/pdf/2502.04095)：提供了RAG与微调的比较分析和实证研究", "https://blog.csdn.net/qq_41187124/article/details/155943887)：详细介绍了微调与RAG的核心对比和选择考量因素", "https://www.ud.hk/sc/blogs/insight/article/2026-04-30-rag-vs-finetuning-enterprise", "https://www.zhihu.com/column/c_1502759624458977280)：提供了RAG和微调的技术选型决策树和具体场景选择指南" ], "citationUrls": [ "https://arxiv.org/pdf/2410.17783", "https://arxiv.org/pdf/2502.04095", "https://blog.csdn.net/qq_41187124/article/details/155943887", "https://www.ud.hk/sc/blogs/insight/article/2026-04-30-rag-vs-finetuning-enterprise", "https://www.zhihu.com/column/c_1502759624458977280" ], "targetCited": false, "brandMentioned": false }, { "provider": "gemini_google_search_grounding", "model": "gemini-2.5-flash", "runIndex": 1, "observedAt": "2026-07-20T23:35:22Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 16, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://arxiv.org/pdf/2510.00386", "https://cloud.google.com/blog/products/ai-machine-learning/optimizing-rag-retrieval", "https://cloud.google.com/use-cases/fine-tuning-ai-models", "https://cohere.com/blog/rag-vs-fine-tuning", "https://docs.nvidia.com/learning/physical-ai/getting-started-with-isaac-sim/latest/synthetic-data-generation-for-perception-model-training-in-isaac-sim/05-fine-tuning-and-validating-an-ai-perception-model.html", "https://medium.com/@bijit211987/when-to-apply-rag-vs-fine-tuning-90a34e7d6d25", "https://medium.com/@candemir13/fine-tuning-vs-rag-a-decision-framework-for-practitioners-7c26cba89768", "https://openreview.net/challenge?redirect=%2Fforum%3Fid%3DfWHd3yYicX", "https://www.actian.com/blog/databases/should-you-use-rag-or-fine-tune-your-llm/", "https://www.braintrust.dev/articles/what-is-rag-evaluation", "https://www.databricks.com/blog/rag-vs-fine-tuning", "https://www.databricks.com/blog/what-is-fine-tuning", "https://www.evidentlyai.com/llm-guide/rag-evaluation", "https://www.meilisearch.com/blog/rag-evaluation", "https://www.oracle.com/artificial-intelligence/generative-ai/retrieval-augmented-generation-rag/rag-fine-tuning/", "https://www.patronus.ai/llm-testing/rag-evaluation-metrics" ], "targetCited": false, "brandMentioned": false }, { "provider": "gemini_google_search_grounding", "model": "gemini-2.5-flash", "runIndex": 2, "observedAt": "2026-07-20T23:35:22Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 10, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://arxiv.org/abs/2401.08406", "https://cohere.com/blog/rag-vs-fine-tuning", "https://dev.to/gde03/fine-tuning-and-rag-what-a-dozen-failed-experiments-taught-me-4apb", "https://docs.aws.amazon.com/prescriptive-guidance/latest/retrieval-augmented-generation-options/rag-vs-fine-tuning.html", "https://galileo.ai/blog/optimizing-llm-performance-rag-vs-finetune-vs-both", "https://unstructured.io/insights/rag-vs-fine-tuning-your-guide-to-ai-optimization", "https://www.actian.com/blog/databases/should-you-use-rag-or-fine-tune-your-llm/", "https://www.coreweave.com/blog/pretraining-vs-fine-tuning-vs-rag-whats-best-for-your-ai-project", "https://www.databricks.com/blog/rag-vs-fine-tuning", "https://www.redhat.com/en/topics/ai/rag-vs-fine-tuning" ], "targetCited": false, "brandMentioned": false }, { "provider": "gemini_google_search_grounding", "model": "gemini-2.5-flash", "runIndex": 3, "observedAt": "2026-07-20T23:35:22Z", "searchPerformed": true, "attributionComplete": true, "rawCitationStringCount": 10, "rawCitationStringsSameAsCitationUrls": true, "citationUrls": [ "https://arxiv.org/pdf/2505.15179", "https://community.ibm.com/community/user/blogs/wendy-munoz/2025/11/28/rag-vs-fine-tuning-best-practices-using-the-ibm-ai", "https://dev.to/gde03/fine-tuning-and-rag-what-a-dozen-failed-experiments-taught-me-4apb", "https://docs.aws.amazon.com/prescriptive-guidance/latest/retrieval-augmented-generation-options/rag-vs-fine-tuning.html", "https://galileo.ai/blog/optimizing-llm-performance-rag-vs-finetune-vs-both", "https://learn.microsoft.com/en-us/azure/developer/ai/augment-llm-rag-fine-tuning", "https://medium.com/@eugenetan_91090/rag-vs-fine-tuned-models-experimenting-with-llm-effectiveness-64c8098d5a0f", "https://medium.com/@vrajdcs/rag-and-fine-tuning-implementation-best-practices-for-scalable-and-accurate-genai-solutions-e038ea957512", "https://redis.io/blog/10-techniques-to-improve-rag-accuracy/", "https://www.databricks.com/blog/rag-vs-fine-tuning" ], "targetCited": false, "brandMentioned": false } ], "crossProviderHostnameOverlap": [ { "providers": [ "qwen_dashscope_web_search", "kimi_builtin_web_search" ], "sharedHostnames": [ "zhuanlan.zhihu.com" ] }, { "providers": [ "qwen_dashscope_web_search", "gemini_google_search_grounding" ], "sharedHostnames": [] }, { "providers": [ "kimi_builtin_web_search", "gemini_google_search_grounding" ], "sharedHostnames": [ "arxiv.org" ] } ], "limitationsZh": [ "这是一个问题、一个语言、一次短时间窗口和每个平台三次运行的便利样本，不能估计平台总体引用概率。", "模型、搜索索引、位置、账号、服务配置和时间变化都可能改变结果。", "原始字符串与规范化URL的差异来自公开规则，不回写或覆盖历史原始观察。", "引用出现不证明来源内容正确、被答案实际吸收、排名靠前或带来访问。", "主机名出现次数不能替代页面级质量审核；同一组织的不同子域名按不同主机名计数。", "本数据集不公开模型回答正文或第三方页面内容，也不对引用来源作质量背书。", "目标页面在观察时尚未公开；目标引用为零是发布前基线，不应用来比较其他域名的可见性。" ], "internalArtifactPaths": [ "geo/reports/2026-07-20-rag-vs-fine-tuning-prepublication-baseline.json" ], "title": "千问、Kimi、Gemini 的 RAG 与微调选型引用样本（2026-07）", "description": "同一中文无品牌问题在千问、Kimi、Gemini 各运行三次后的引用计数、主机名分布、平台交集、目标引用与方法限制。", "url": "https://course.dragonai.tech/datasets/ai-search-rag-finetuning-citation-sample-2026-07.json", "creator": "烛龙智元内容研究组" }