语义视图：跨语义查询的 LLM 标签认证复用
Semantic Views: Certified Reuse of LLM Labels Across Semantic Queries
Oktie Hassanzadeh · Dharmashankar Subramanian
中文摘要
语义查询引擎使用大语言模型评估自然语言过滤条件，每次评估都是一次代价高昂的预言机调用。降低这些调用次数的优化器面临两难：级联方案信赖廉价代理模型时在许多查询上达不到精度与召回率目标；具有有限样本保证的方法节省调用很少；所有方法在查询结束后都会丢弃预言机的标签。提出 SemViews，一种认证语义过滤器：先依据无需预言机调用的信号（先前查询留下的语义视图标签、本地嵌入与辅助模型打分）对元组排序，再利用无论如何都要付出的代价获取的标签精排顺序，最后用新的分层预言机抽样证明答案的正确性。该答案对任意工作负载与目录都以至少 1−δ 的概率同时满足用户精度与召回率目标，因此陈旧或误导性的视图至多增加预言机调用，却不会损害准确性。在四个重放 725,000 条已记录预言机标签的工作负载上，SemViews 在 99.63% 的查询上同时满足两个目标，节省 28%–45% 的预言机调用；而研究中唯一具有相同保证的先前方法需要 1.32x–1.71x 的调用次数。当同一问题以不同措辞重现时，视图进一步减少 27%–36% 的调用，且至少 99.8% 的此类查询同时满足双目标；相比之下，相似性缓存与基于蕴涵的复用方法仅在 17%–47% 的查询上达成目标。
关键要点
- 01问题：语义查询中每次 LLM 过滤都是昂贵预言机调用，现有优化器要么牺牲精度召回，要么节省有限，且查询结束即丢弃标签
- 02方法：SemViews 用零成本信号（复用视图、嵌入、辅助打分）预排序，再用必需调用精排，并以分层抽样提供概率保证
- 03结果：在 725,000 条标签的重放中，99.63% 查询同时满足精度与召回目标，节省 28%–45% 调用，且对重述查询再省 27%–36%
- 04局限：保证建立在概率 1−δ 之上，对陈旧或误导性视图至多增加调用成本，认证保证需足够分层抽样支撑
解读
尚无解读。
原始英文摘要
arXiv:2610.12768v1 Announce Type: cross Abstract: Semantic query engines evaluate natural-language filters with a large language model, and every evaluation is a costly oracle call. Optimizers that reduce these calls face a dilemma: cascades that trust a cheap proxy miss their precision and recall targets on many queries, methods with finite-sample guarantees save little, and all of them discard the oracle's labels when the query ends. We present SemViews, a certified semantic filter that orders tuples by signals that cost no oracle calls (the labels that earlier queries left behind as semantic views, a local embedding, and a helper model's score), refines that order with labels it must pay for anyway, and certifies its answer with fresh stratified oracle samples. The answer meets user precision and recall targets with probability at least $1-\delta$ for every workload and every catalog, so a stale or misleading view can cost oracle calls but never accuracy. On four workloads replayed against 725,000 recorded oracle labels, SemViews meets both targets on 99.63% of queries and saves 28%-45% of the oracle calls, whereas the only prior method in our study with the same guarantee needs 1.32x-1.71x as many calls. When a question returns in different words, views cut its calls by a further 27%-36%, with both targets met on at least 99.8% of such queries; similarity caches and entailment-based reuse meet them on only 17%-47%.