跳转至

市场目录的自动研究:从传统表单到AI原生匹配

文章背景与核心概要

双边服务市场正在经历一场深刻的变革,从传统的、确定性的请求表单录入系统,迅速向AI原生、基于概率的匹配模式转型。大语言模型(LLM)的兴起使得平台能够直接从用户的自然语言中推断出其潜在的意图、偏好和约束条件。然而,这种转变也带来了一个根本性的挑战:平台必须同步重构那些支撑底层搜索、定价和匹配机制的供给端偏好分类体系,以适应基于意图的交互方式。

本文介绍了一种在一家美国大型消费服务市场(截至2026年4月涵盖132个职业)中投入生产的自动研究循环(autoresearch loop)系统。该系统没有采用僵化的全局层级结构,而是将每个职业视为独立的生成问题,通过迭代式的“提出-评估-保留”优化循环来动态生成分类标签。其中包含基于重新校准的六维度评分框架(LLM-as-Judge)、由7个不同角色组成的批评家小组(提供加权惩罚而非硬性一票否决),以及一个用于将传统历史表单问题与新生成分类相映射的对等映射阶段。这一创新不仅提高了匹配精度,还为覆盖率分析和人工质量保证提供了可靠的信号。


📌 Summary

Two-sided service marketplaces are rapidly transitioning from deterministic request-form intake systems to AI-native, probabilistic matching. By utilizing Large Language Models (LLMs) to infer latent user intent, preferences, and constraints from natural language, platforms must simultaneously regenerate the underlying provider-side preference taxonomies that drive search, pricing, and matching.

This paper introduces an autoresearch loop deployed in production at a major U.S. consumer services marketplace (spanning 132 occupations as of April 2026). Instead of relying on a rigid global hierarchy, the system treats each occupation as an independent generation problem using iterative propose-evaluate-keep refinement cycles: * LLM-as-Judge Framework: Scores candidate tag sets via a recalibrated six-rubric evaluation system. * 7-Critic Persona Panel: Contributes weighted penalties to adjust scores without enforcing hard vetoes. * Parity-Mapping Stage: Bridges legacy request-form Q&A pairs back to the newly generated taxonomy by inferring original intent rather than relying on literal translation, simultaneously providing a signal for coverage and human quality assurance.

双边服务市场正在迅速从确定性的请求表单录入系统过渡到AI原生的概率匹配。通过利用大语言模型(LLM)从自然语言中推断用户的潜在意图、偏好和约束,平台必须同时重新生成底层驱动搜索、定价和匹配的供给端偏好分类学。

本文介绍了一个在一家美国大型消费服务市场(截至2026年4月跨越132个职业)投入生产的自动研究循环。该系统没有依赖僵化的全局层级结构,而是通过迭代的“提出-评估-保留”精炼循环,将每个职业视为一个独立的生成问题: * LLM裁判框架(LLM-as-Judge Framework): 通过重新校准的六项评估系统对候选标签集进行评分。 * 7位批评家角色小组(7-Critic Persona Panel): 提供加权惩罚以调整分数,而不实施硬性一票否决。 * 对等映射阶段(Parity-Mapping Stage): 通过推断原始意图而非依赖字面翻译,将传统的请求表单问答对桥接到新生成的分类学中,同时提供覆盖率信号和人工质量保证接口。


Field Details
Primary Subject Artificial Intelligence (cs.AI)
DOI 10.48550/arXiv.2609.00274
License Creative Commons Attribution 4.0
Direct Access View PDF | HTML Version | TeX Source
Field Details
Primary Subject Artificial Intelligence (cs.AI)
DOI 10.48550/arXiv.2609.00274
License Creative Commons Attribution 4.0
Direct Access View PDF | HTML Version | TeX Source

🔍 Abstract

Two-sided service marketplaces are moving from deterministic request-form intake to AI-native probabilistic matching, enabled by large language models (LLMs) that infer intent, preferences, and latent constraints from natural language. Relying on inferred intent rather than fixed-form fields forces these platforms to regenerate the provider-side preference taxonomy underwriting matching, search, and pricing: attributes interpretable to service providers while remaining a useful signal for marketplace decisions. We present an autoresearch loop that generates this taxonomy, one occupation at a time, and has been deployed in production at a major U.S. consumer services marketplace since April 2026, spanning 132 occupations. Instead of one global hierarchy, the loop treats each occupation as an independent generation problem and runs iterative propose-evaluate-keep refinement cycles. Each candidate tag set is scored by a recalibrated six-rubric LLM-as-judge framework, and a 7-critic panel of distinct personas contributes weighted penalties to an adjusted score, with no hard vetoes. A separate parity-mapping stage maps legacy request-form Q&A pairs back to the generated taxonomy, yielding both a coverage signal and an interface for human quality assurance; it does so by first inferring the provider attribute each legacy question was meant to measure, rather than translating questions to tags literally.

双边服务市场正在从确定性的请求表单录入转向AI原生的概率匹配,大语言模型(LLM)使这一转变成为可能,它能够从自然语言中推断意图、偏好和潜在约束。依赖推断出的意图而非固定表单字段,迫使这些平台重新生成支持匹配、搜索和定价的供给端偏好分类学:这些属性既可被服务提供者理解,又能保持作为市场决策有用信号的特性。我们提出了一种自动研究循环,它逐个职业地生成这种分类学,自2026年4月起已在一家美国大型消费服务市场的生产环境中部署,涵盖132个职业。该循环没有采用单一的全局层级结构,而是将每个职业视为独立的生成问题,并运行迭代的“提出-评估-保留”精炼循环。每个候选标签集由重新校准的六维度LLM裁判框架进行评分,由不同角色组成的7位批评家小组为调整后的分数贡献加权惩罚,且不设硬性否决。一个独立的对等映射阶段将传统的请求表单问答对映射回生成的分类学中,同时产生覆盖率信号和人工质量保证接口;其实现方式是首先推断每个传统问题旨在测量的服务商属性,而不是将问题字面翻译为标签。