[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83022-en":3,"doc-seo-83022-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83022,7971461740909,"Levi","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","InfluMatch Frontier-Quality KOL Search at 4B-Model Cost","InfluMatch matches influencers (KOLs) to free-form, multi-part Thai marketing criteria where traditional keyword search lacks semantic alignment and frontier-LLM prompting is too costly. The method uses a low-cost three-stage cascade—retrieval, rerank, and reason—implemented entirely with small open-weight 4B models. Dense retrieval narrows candidates, a pointwise reranker scores candidates via a single-token Yes log-probability, and a reasoner grades per-criterion with Thai rationales. On an 11-query evaluation, the cascade reaches 94.1% P@5 while reducing token spend versus full candidate reasoning, matching frontier performance at a fraction of serving cost.","arXiv :2607 .05968v 1 [ cs .CL] 7 Jul 2026  \nInfluMatch: Frontier-Quality KOL Search at 4B-Model Cost  \nKrittanon Kaewtawee, Petmongkon Pornpichitsuwan, Natchaya Temyingyong, Nutnicha Laplamoon, Wachiravit Modecrua,  \nKrittin Pachtrachai, Touchapon Kraisingkorn  \nAmity AI Holdings Co.,Ltd.  \n{krittanon, petmongkon, natchaya, nutnicha.lap, wachiravit, krittin, [touchapon}@amity.co](touchapon}@amity.co)  \nAbstract  \nMatching 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 → rerank → 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:  \nreasoning over a filtered top-10 halves token spend versus reasoning over all 50 while scoring  \n14 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 ∼35 × fewer output tokens and serving a 50-KOL query in ∼20s 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.  \n1 Introduction  \nInfluencer marketing allocates a large share of brand budgets, and its central operational problem is matching: given a campaign, find the Key Opinion Leaders (KOLs) whose audience, content, and voice fit the brief. In the Thai market, a marketer’s intent is naturally expressed as free-form, multi-part criteria—“a female food reviewer with a playful tone whose audience skews young”—while the supply side is a long tail of creators whose relevance must be inferred from heterogeneous, largely Thai signals (bios, video transcripts, audience profiles) rather than read off a clean attribute table.  \nWhy keyword and structured search fall short. The incumbent approach treats matching as lookup over a scraped KOL database filtered by keywords and hard attributes. This fails on three counts: it is lexical, not semantic (a “playful food reviewer” query misses a KOL whose bio never says so but whose content plainly fits); the signals are fragmented across profile fields, transcripts, and engagement metrics, so a query can satisfy each field in isolation yet return a poor overall match; and the deciding criteria are dynamic, campaign-specific predicates that a static schema cannot  \nanticipate. Our evaluation confirms the ceiling of pure recall: dense retrieval alone barely separates from a random baseline on end-to-end precision (54.5% vs. 54.0% P@5, §6.3) .  \nApproach. We present InfluMatch, a three-stage pipeline—retrieval → rerank → reason —that turns marketer criteria into a ranked KOL shortlist with per-criterion scores and Thai rationales. Stage 1 retrieves a top-50 candidate set from a vector index; Stage 2 applies a pointwise 4B LLMreranker that keeps the top 10; Stage 3 scores each shortlisted KOL per criterion and combines the scores by an even sum. To supervise the learned stages we build an evidence-grounded, multiobjective human-labeled corpus over synthetic-but-realistic Thai briefs—pointwise ordinal scores, binary pass/fail verdicts, and pa","cbCair57RmD7m3WR","https://ap.wps.com/l/cbCair57RmD7m3WR","pdf",757979,2,1,15,"English","en",105,"# Abstract\n# 1 Introduction\n## 1.1 Contributions\n# 2 Background and Related Work","[{\"question\":\"What problem does InfluMatch address in Thai influencer marketing?\",\"answer\":\"It addresses matching KOLs to free-form, multi-part Thai marketing criteria, where marketer intent is not easily captured by static attributes and existing keyword search is not semantically aligned.\"},{\"question\":\"How does the three-stage InfluMatch pipeline work?\",\"answer\":\"Stage 1 retrieves a top-50 candidate set, Stage 2 reranks candidates with a pointwise 4B model that keeps the best 10, and Stage 3 scores the shortlist per criterion and combines them into a final ranked shortlist with Thai rationales.\"},{\"question\":\"Why is pairwise fine-tuning for the reranker emphasized?\",\"answer\":\"Pairwise SimPO-tuning yields the strongest end-to-end results: it matches the frontier baseline’s best-pick accuracy, while other fine-tuning choices improve offline metrics but degrade end-to-end ranking.\"}]",1784184709,38,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"influmatch-frontier-quality-kol-search-at-4b-model-cost","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/influmatch-frontier-quality-kol-search-at-4b-model-cost/83022/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-24","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does InfluMatch address in Thai influencer marketing?","Question",{"text":75,"@type":76},"It addresses matching KOLs to free-form, multi-part Thai marketing criteria, where marketer intent is not easily captured by static attributes and existing keyword search is not semantically aligned.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the three-stage InfluMatch pipeline work?",{"text":80,"@type":76},"Stage 1 retrieves a top-50 candidate set, Stage 2 reranks candidates with a pointwise 4B model that keeps the best 10, and Stage 3 scores the shortlist per criterion and combines them into a final ranked shortlist with Thai rationales.",{"name":82,"@type":73,"acceptedAnswer":83},"Why is pairwise fine-tuning for the reranker emphasized?",{"text":84,"@type":76},"Pairwise SimPO-tuning yields the strongest end-to-end results: it matches the frontier baseline’s best-pick accuracy, while other fine-tuning choices improve offline metrics but degrade end-to-end ranking.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]