[Submitted on 7 Jul 2026]

标题：InfluMatch：以4B模型成本实现前沿品质的KOL搜索

查看论文 PDF，标题为 InfluMatch: Frontier-Quality KOL Search at 4B-Model Cost，作者 Krittanon Kaewtawee 等 6 人

查看 PDF HTML (实验性)

摘要：将网红（KOL）匹配到自由形式、多部分的泰语营销标准，目前要么通过对结构化档案的关键词搜索，但会遗漏语义适配性；要么通过向每个候选者提示前沿大语言模型，虽准确但速度慢且昂贵。我们提出了 InfluMatch，一种低成本的三级级联——检索 → 重排 → 推理——完全由小型开放权重模型构建：稠密检索返回 50 个候选，一个 4B 逐点重排模型通过单个“是”标记的对数概率打分并保留 10 个，然后一个 4B 推理模型依据评分标准对候选短名单进行逐标准评分，并给出泰语推理。该级联以成本为设计重点：对过滤后的前 10 名进行推理，相比对所有 50 名推理，token 消耗减半，同时得分高出 14 个百分点。在包含所有 50 个候选标注的 11 个查询集上，与人类相关性标签进行端到端比较，整个级联达到了 94.1% 的 P@5，而纯检索基线接近随机；它与前沿模型 Kimi-K2.6（91.8%）性能相当，但输出的 token 数减少约 35 倍，且在单块 A100 上处理一个 50 个 KOL 的查询仅需约 20 秒。值得注意的是，唯一有效的微调是成对微调：经 SimPO 调优的重排模型在最佳选择准确率（78.0 EM）上达到了前沿基线的水平，而对推理模型进行逐点分标准标注的微调虽然提升了离线分数，却损害了端到端排名——我们将这种反转归因于绝对标注任务的设计——使得未经微调的基础模型成为最强的部署推理器。最终成果是一个可部署、可解释的 KOL 搜索系统，其服务成本仅为前沿模型的一小部分。

提交历史

来自: Krittanon Kaewtawee [

查看邮件

]

[v1] 周二，2026年7月7日 08:04:05 UTC (1,435 KB)

[Submitted on 7 Jul 2026]

Title:InfluMatch: Frontier-Quality KOL Search at 4B-Model Cost

View a PDF of the paper titled InfluMatch: Frontier-Quality KOL Search at 4B-Model Cost, by Krittanon Kaewtawee and 6 other authors

View PDF HTML (experimental)

Abstract:Matching influencers (KOLs) to free-form, multi-part Thai marketing criteria is today served either by keyword search over structured profiles, which misses semantic fit, or by prompting frontier LLMs over every candidate, which is accurate but slow and expensive. We present InfluMatch, a low-cost three-stage cascade -- retrieval $\rightarrow$ rerank $\rightarrow$ reason -- built entirely from small open-weight models: dense retrieval returns 50 candidates, a 4B pointwise reranker scores each by the log-probability of a single Yes token and keeps 10, and a 4B reasoner grades the shortlist per criterion on a rubric with a Thai rationale. The cascade is designed for cost: reasoning over a filtered top-10 halves token spend versus reasoning over all 50 while scoring 14 points higher. End-to-end against human relevance labels on an 11-query set with all 50 candidates labeled, the full cascade reaches 94.1% P@5, versus a retrieval-only baseline near random; it matches the frontier model Kimi-K2.6 (91.8%) while emitting ${\sim}35\times$ fewer output tokens and serving a 50-KOL query in ${\sim}20$ s on one A100. Notably, the only fine-tuning that pays off is pairwise: a SimPO-tuned reranker matches the frontier baseline's best-pick accuracy (78.0 EM), whereas fine-tuning the reasoner on pointwise per-criterion labels improves offline scores yet degrades end-to-end ranking -- an inversion we trace to the design of the absolute labeling task -- leaving the untuned base model as the strongest deployed reasoner. The result is a deployable, explainable KOL search system at a small fraction of frontier serving cost.

Submission history

From: Krittanon Kaewtawee [

view email

]

[v1]

Tue, 7 Jul 2026 08:04:05 UTC (1,435 KB)