大模型推荐系统常自说自话，却对自己的错误毫无察觉。
Do LLM Recommenders Know When They're Hallucinating? Auditing Confidence Calibration in Catalog Faithfulness
- 用置信度校准评估四大厂商的零样本推荐模型
- 亚马逊玩具推荐幻觉率高达49%-61%，而电影推荐仅0.6%-2.7%
- 模型置信度恒定，与真实命中率无关，需改用锚定目录的提问方式
针对零样本大语言模型推荐系统在Top-K推荐中频繁生成不在目标目录中的项目（即幻觉）的问题，本文联合审计了四个独立厂商（Mistral Large、Llama-3.3-70B、GPT-OSS-120B、Claude Sonnet 4.6）的推荐结果，覆盖三个数据集（MovieLens-25M、Amazon Reviews 2023 Toys、Yelp Open Dataset），并按物品流行度分层。研究发现：幻觉率高度依赖目录，亚马逊玩具目录达49.3%-61.0%，而电影目录仅为0.6%-2.7%；各模型置信度几乎恒定，不随目录变化，导致7个组合低估、5个高估错误，所有模型在电影目录上均低估，在玩具目录上均高估。这反映“直接提问”会诱发通用质量评分而非目录匹配概率。通过201次人工标注验证，采用净偏差最小的字符串匹配器后，误差从+0.144降至-0.040。此外，基于置信度的合规性弃权阈值最多仅降低1.65个百分点的幻觉率，因该通道无法区分正确项与幻觉项。建议未来审计应同时报告校准性、验证匹配器，并使用目录锚定的提问策略。
原文摘要 · Abstract (English)
LLM recommenders for top-K item suggestion regularly emit titles outside the target catalog. Prior audits report a binary out-of-domain rate; none ask whether the model knew. We jointly audit hallucination rate (OOD@10) and verbalized-confidence calibration (ECE, Brier, reliability) for four zero-shot LLM recommenders from four independent vendors (Mistral Large, Llama-3.3-70B, GPT-OSS-120B, Claude Sonnet 4.6), not grounded or fine-tuned systems, across three catalogs (MovieLens-25M, Amazon Reviews 2023 Toys, Yelp Open Dataset), stratified by item popularity. Measuring catalog membership is itself the hard part: on identical outputs the reported rate moves by an order of magnitude with the string matcher used, and F1 cannot separate the candidates. We validate the instrument against 201 human judgments and select on net bias, where the adopted one is off by -0.040 against +0.144 for the common fuzzy rule. Hallucination is then strongly catalog-dependent (0.6-2.7% on MovieLens, 11.6-38.7% on Yelp, 49.3-61.0% on Amazon Toys). Each model holds a near-constant confidence level barely responsive to the catalog, while the catalog-hit rate swings 60 points, so the sign of the error is set by where a model's constant lands against a catalog's accuracy: 7 of the twelve cells are under-confident and 5 over-confident, all four under-confident on MovieLens, all four over-confident on Amazon Toys. We read this as an elicitation mismatch: "Just Ask" elicits a generic quality rating, not a catalog-membership probability. A conformal abstention threshold over verbalized confidence changes hallucination by at most 1.65 pp across four alpha levels, because the channel cannot separate correct items from hallucinations. We recommend that audits report calibration alongside OOD, validate the matcher producing the OOD number, and use catalog-anchored elicitation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。