[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-195422-105":53,"doc-detail-195422-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","250702020-table-schema-matching-research-milestones","2507.02020 - Table Schema Matching Research Milestones","","Research traces the evolution of table and schema matching methods across multiple years, from early ML-based schema matching to modern deep learning approaches for table detection and complex structure recognition. It details a template-driven column-to-target-attribute mapping workflow that combines schema-based and instance-based metrics with weighted hybrid scoring. The final assignment uses a Hungarian Matching Algorithm with a threshold, and results report improvements in F1, precision, recall, compactness, and null-rate under optimized parameters and ALITE variants.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":46,"@type":70,"position":76},"https://docshare.wps.com/template/paper-templates/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/250702020-table-schema-matching-research-milestones/195422/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/250702020-table-schema-matching-research-milestones/195422.png","ImageObject",442,249,{"name":88,"@type":89},"Maya Linwood","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-10-03","2026-09-03",true,{"@type":98,"interactionType":99,"userInteractionCount":76},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"How does the column-to-target-attribute mapping decide matches?","Question",{"text":108,"@type":109},"Each extracted table column header is scored against every target schema attribute using a weighted hybrid of schema-based and instance-based metrics. Assignments below the threshold are set to negative infinity before optimization.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"Why is the Hungarian Matching Algorithm used?",{"text":113,"@type":109},"After computing hybrid scores, the Hungarian Matching Algorithm finds a globally optimal assignment between column headers and target attributes, improving over greedy matching.",{"name":115,"@type":106,"acceptedAnswer":116},"What performance metrics are reported for different matching methods?",{"text":117,"@type":109},"Results include F1, precision, recall, compactness (number of columns), and null percentage. Comparisons show gains under optimized hybrid parameters and weight strategies, as well as ALITE variants.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},195422,1790317481,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":45,"category_name":46,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":76,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":40,"language":135,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":136,"faqs":137,"seo_title":138,"seo_description":61,"update_tm":139,"read_time":140},962084928432,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","| Year | Milestone | Description |\n| --- | --- | --- |\n| 2001 | LSD | Early ML-based schema matching. |\n| 2002 | Cupid / COMA /\u003Cbr>Similarity Flooding | Linguistic and structural matchers (Cupid), graph-based matching (SF), and composite matchers (COMA) . |\n| 2005 | COMA++ | Reuse of prior mappings |\n| 2017 | TableNet | CNN-based table detection |\n| 2018 | Table Union Search | Matching union-able tables at scale |\n| 2019 | GraphTSR | Complex table structure recognition |\n| 2020 | Valentine | Benchmarking framework for schema matching algorithms |\n| 2021 | PubTables-1M | One million annotated tables; boosted deep table extraction models |\n| 2022 | ALITE | Full Disjunction for schema integration |\n| 2023 | DeepJoin | BERT-based semantic join-ability prediction |\n| 2024 | Tabular Augmentation | Use of generative AI for augmenting tabular training data |\n| 2025 | Retrieve, Merge, Predict | Automated table enrichment from data lakes |\n| 2025 | Soperla | Empirical study: highlights manual bottlenecks |\n| 2025 | This Research | Template-based schema matching for multi-layout PDFs (tenany schedules) |\n\n| Source Office Area Annual Rent Commencement |\n| --- |\n| Total 0.95 0.94 0.12\u003Cbr>Rent 0.93 0.60 0.17\u003Cbr>Start 0.20 0.30 0.98 |\n| A greedy algorithm would result in a total score of 0.95 + 0.6 + 0.98 = 2. 53, whereas the globally optimal assignment achieves a total of 0.93 + 0.94 + 0.98 = 2.85.\u003Cbr>3.2.4 Overview and Pseudo-Code. The following Algorithm 1 outlines the full column-to-target-attribute matching process described above. For each extracted table column it is scored against every target attribute in the template schema. The matching is done using a combination of schema-based and instance-based metrics. The final assignment to target attributes is determined via the Hungarian Matching Algorithm with a threshold. |\n| Algorithm 1: Column-to-Target-Schema |\n| Input: DataFrame 􀀩 from CSV with headers 􀀝 ; target schema attributes 􀀨; metric weights 􀁆 􀀼; alpha 􀁕; threshold 􀁜\u003Cbr>Output: Mapping 􀀢 : 􀀝 → 􀀨\u003Cbr>1 foreach header ℎ ∈ 􀀝 do\u003Cbr>2 foreach attribute 􀀰 ∈ 􀀨 do\u003Cbr>3 􀁂schema ←Í􀀼∈􀀢schema 􀁆 􀀼 · 􀀼(ℎ, 􀀰) ;\u003Cbr>4 􀁂instance ←Í􀀼∈􀀢instance 􀁆 􀀼 · 􀀼 (􀀩 [ℎ], 􀀰) ;\u003Cbr>5 􀁂 hybrid ← 􀁕 · 􀁂schema + (1 − 􀁕 ) · 􀁂instance ;\u003Cbr>6 if 􀁂 hybrid ≥ 􀁜 then\u003Cbr>7    􀀘 [ℎ, 􀀰] ← 􀁂 hybrid ;\u003Cbr>8 else\u003Cbr>9        􀀘 [ℎ, 􀀰] ← −∞ ;\u003Cbr>10 (rows, cols) ← Hungarian Algorithm(−􀀘) ;\u003Cbr>11 􀀢 ← {ℎ 􀀸 ↦→ 􀀰 􀀹 | (􀀸, 􀀹 ) ∈ (rows, cols) ∧ 􀀘 [􀀸, 􀀹 ] ≠ −∞} ;\u003Cbr>12 return 􀀢\u003Cbr>|\n\n| Method | F1 | Precision | Recall | Compactness (\\# columns) | Null % |\n| --- | --- | --- | --- | --- | --- |\n| Hybrid (baseline: 􀁕 = 0.50, 􀁜 = 0. 50) | 0.778 | 0.789 | 0.767 | 17 | 47.8% |\n| Hybrid (optimized params: 􀁕 = 0.75, 􀁜 = 0. 51) | 0.800 | 0.795 | 0.806 | 17 | 50.1% |\n| Hybrid (optimized params + weights) | 0.875 | 0.913 | 0.840 | 17 | 45.9% |\n| Hybrid (optimized weights only, 􀁕 = 0.50, 􀁜 = 0. 50) | 0.881 | 0.926 | 0.840 | 17 | 45.7% |\n| ALITE Variants |  |  |  |  |  |\n| ALITE (all-mpnet-base-v2)* | 0.699 | 1.000 | 0.538 | 89 | 75.9% |\n| ALITE (sentence-t5-base)* | 0.712 | 1.000 | 0.553 | 88 | 75.6% |\n\n\n| Type | Schema Avg | Instance Avg | Gap |\n| --- | --- | --- | --- |\n| STRING | 0.1500 | 0.2579 | 0.108 |\n| DECIMAL | 0.1143 | 0.292 | 0.178 |\n| DATE | 0.1000 | 0.356 | 0.256 |","cbCaibXadcDZB1SX","https://ap.wps.com/l/cbCaibXadcDZB1SX","pdf",731985,"English","# Overview and Pseudo-Code\n## Column-to-Target-Schema Algorithm\n## Evaluation Results\n## Metric Breakdown (STRING/DECIMAL/DATE)","[{\"question\":\"How does the column-to-target-attribute mapping decide matches?\",\"answer\":\"Each extracted table column header is scored against every target schema attribute using a weighted hybrid of schema-based and instance-based metrics. Assignments below the threshold are set to negative infinity before optimization.\"},{\"question\":\"Why is the Hungarian Matching Algorithm used?\",\"answer\":\"After computing hybrid scores, the Hungarian Matching Algorithm finds a globally optimal assignment between column headers and target attributes, improving over greedy matching.\"},{\"question\":\"What performance metrics are reported for different matching methods?\",\"answer\":\"Results include F1, precision, recall, compactness (number of columns), and null percentage. Comparisons show gains under optimized hybrid parameters and weight strategies, as well as ALITE variants.\"}]","2507.02020 - Table Schema Matching Research Milestones | PDF",1788448351,6]