{"id":1316,"date":"2024-10-14T20:00:20","date_gmt":"2024-10-14T12:00:20","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=1316"},"modified":"2024-10-14T20:00:20","modified_gmt":"2024-10-14T12:00:20","slug":"202410-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-robustness-of-structured-data-extraction-from-in-plane-rotated-documents-using-multi-modal-large-language-models-llm","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202410-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-robustness-of-structured-data-extraction-from-in-plane-rotated-documents-using-multi-modal-large-language-models-llm\/","title":{"rendered":"202410 \u8bba\u6587\u7814\u8bfb-Robustness of Structured Data Extraction from In-Plane Rotated Documents Using Multi-Modal Large Language Models (LLM)"},"content":{"rendered":"\n

\u4f5c\u8005\uff1a<\/strong> Anjanava Biswas , Wrick Talukdar<\/p>\n\n\n\n

\u671f\u520a\uff1a<\/strong> Journal of Artificial Intelligence Research (CCF-B)<\/p>\n\n\n\n

\u65f6\u95f4\uff1a<\/strong> 2024<\/p>\n\n\n\n

1.\u80cc\u666f<\/h2>\n\n\n\n

\u00a0\u00a0\u00a0 \u968f\u7740\u591a\u6a21\u6001\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u4efb\u52a1\u4e2d\u5c55\u73b0\u51fa\u663e\u8457\u6027\u80fd\uff0c\u5176\u5728\u6587\u6863\u6570\u636e\u63d0\u53d6\u4e2d\u7684\u5e94\u7528\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\u3002\u7136\u800c\uff0c\u73b0\u5b9e\u4e16\u754c\u4e2d\u626b\u63cf\u6587\u6863\u5e38\u5e38\u5b58\u5728\u5e73\u9762\u65cb\u8f6c\uff08\u5373\u503e\u659c\uff09\uff0c\u8fd9\u4e00\u5e38\u89c1\u95ee\u9898\u4f1a\u663e\u8457\u5f71\u54cd\u6a21\u578b\u7684\u51c6\u786e\u6027\u3002\u4ee5\u76ee\u524d\u6bd4\u8f83\u5148\u8fdb\u7684\u4e09\u4e2a\u5927\u8bed\u8a00\u6a21\u578b\u4e3a\u4f8b\uff08Anthropic Claude V3 Sonnet\u3001GPT-4-Turbo\u548cLlava:v1.6\uff09\uff0c\u672c\u6587\u65e8\u5728\u7cfb\u7edf\u7814\u7a76\u5927\u6a21\u578b\u5728\u8bc6\u522b\u7406\u89e3\u6587\u6863\u5173\u952e\u952e\u503c\u5bf9\u4fe1\u606f\u65f6\uff0c\u6587\u6863\u503e\u659c\u5bf9\u5176\u8bc6\u522b\u6548\u679c\u7684\u5f71\u54cd\u3002<\/p>\n\n\n\n

2.\u6838\u5fc3\u5185\u5bb9<\/h2>\n\n\n\n

\u00a0\u00a0\u00a0 \u00a0\u00a0\u00a0\u00a0\u672c\u7814\u7a76\u7cfb\u7edf\u5730\u8bc4\u4f30\u4e86\u6587\u6863\u503e\u659c\u5bf9\u4e09\u4e2a\u5148\u8fdb\u591a\u6a21\u6001\u5927\u6a21\u578b\u63d0\u53d6\u6587\u6863\u4e2d\u5173\u952e\u952e\u503c\u5bf9\u4fe1\u606f\u7684\u6027\u80fd\u5f71\u54cd\u3002\u671f\u671b\u5229\u7528\u591a\u6a21\u6001LLM\u6267\u884c\u7c7b\u4f3cOCR\u7684\u529f\u80fd\uff0c\u4ee5\u8fdb\u884c\u51c6\u786e\u7684\u7ed3\u6784\u5316\u6570\u636e\u63d0\u53d6\u3002\u7814\u7a76\u7684\u4e3b\u8981\u8d21\u732e\u5305\u62ec\uff1a<\/p>\n\n\n\n

\u00a0\u00a0\u00a0 1.\u660e\u786e\u6587\u6863\u504f\u659c\u5bf9\u6240\u6d4b\u8bd5\u7684\u4e09\u79cd\u591a\u6a21\u6001LLM\u7684\u63d0\u53d6\u7cbe\u5ea6\u90fd\u6709\u663e\u8457\u5f71\u54cd\u3002<\/p>\n\n\n\n

    2.\u786e\u5b9a\u4e86\u4e0d\u540c\u6a21\u578b\u7684\u5b89\u5168\u5e73\u9762\u65cb\u8f6c\u89d2\u5ea6\uff08SIPRA\uff09\uff0c\u5373\u6a21\u578b\u80fd\u591f\u7ef4\u6301\u8f83\u9ad8\u63d0\u53d6\u51c6\u786e\u6027\u7684\u65cb\u8f6c\u89d2\u5ea6\u8303\u56f4\u3002<\/p>\n\n\n\n

    3.\u5206\u6790\u4e86\u503e\u659c\u5bf9\u6a21\u578b\u4ea7\u751f\u5e7b\u89c9\uff08\u751f\u6210\u9519\u8bef\u6216\u865a\u6784\u4fe1\u606f\uff09\u7684\u5f71\u54cd\u3002<\/p>\n\n\n\n

    4.\u63d0\u51fa\u4e86\u65b0\u7684\u591a\u6a21\u6001\u67b6\u6784\u548c\u9884\u8bad\u7ec3\u9636\u6bb5\u7684\u503e\u659c\u6280\u672f\uff0c\u4ee5\u589e\u5f3a\u6a21\u578b\u5bf9\u6587\u6863\u503e\u659c\u7684\u9c81\u68d2\u6027\u3002<\/p>\n\n\n\n

3.\u5b9e\u9a8c\u8fc7\u7a0b\u53ca\u7ed3\u679c<\/h2>\n\n\n\n

\uff081\uff09 \u51c6\u5907\u4e00\u4e2a\u5e26\u6709\u57fa\u672c\u4e8b\u5b9e\u952e\u503c\u5bf9\u6ce8\u91ca\u7684\u6587\u6863\u6570\u636e\u96c6\u3002\u5728\u539f\u59cb\u7684\u3001\u672a\u503e\u659c\u7684\u6587\u6863\u56fe\u50cf\u4e2d\u624b\u52a8\u6ce8\u91ca\u952e\u503c\u5bf9\u3002\u8fd9\u4e9b\u57fa\u672c\u771f\u503c\u7528\u4f5c\u8bc4\u4f30LLM\u4ece\u503e\u659c\u6587\u6863\u56fe\u50cf\u4e2d\u63d0\u53d6\u7684\u952e\u503c\u5bf9\u51c6\u786e\u6027\u7684\u53c2\u8003\u3002\u4ee5\u7f16\u8f91\u8ddd\u79bb\uff08Levenshtein\uff09\u6765\u8861\u91cf\u4e24\u4e2a\u5b57\u7b26\u4e32\u4e4b\u95f4\u7684\u5dee\u5f02\u6027\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n


\uff082\uff09\u4f7f\u7528\u56fe\u50cf\u5904\u7406\u6280\u672f\u5bf9\u6587\u6863\u5f15\u5165\u4e0d\u540c\u7a0b\u5ea6\u7684\u503e\u659c\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n


\uff083\uff09\u5c06\u504f\u659c\u7684\u6587\u6863\u4f20\u9012\u7ed9\u6240\u9009\u7684\u591a\u6a21\u6001LLM\uff0c\u5e76\u63d0\u53d6\u952e\u503c\u5bf9\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n


\uff084\uff09\u4f7f\u7528\u7cbe\u786e\u5ea6\u3001\u53ec\u56de\u7387\u548cF1\u5206\u6570\u7b49\u8bc4\u4f30\u6307\u6807\u5c06\u63d0\u53d6\u7684\u952e\u503c\u5bf9\u4e0e\u771f\u5b9e\u6ce8\u91ca\u8fdb\u884c\u6bd4\u8f83<\/p>\n\n\n\n

Claude V3 Sonnet\u6a21\u578b\u5b9e\u9a8c\u7ed3\u679c<\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

GPT-4-turbo\u6a21\u578b\u5b9e\u9a8c\u7ed3\u679c<\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

Llava:v1.6\u6a21\u578b\u5b9e\u9a8c\u7ed3\u679c<\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

\u5bf9\u6bd4\u4e09\u4e2a\u5927\u6a21\u578b\u5bf9\u4e0d\u540c\u503e\u659c\u6587\u6863\u7684\u8bc6\u522b\u7ed3\u679c\u53ef\u4ee5\u5f97\u77e5\uff0c\u4e09\u79cd\u591a\u6a21\u6001LLM\u90fd\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u53d7\u5230\u6587\u6863\u504f\u659c\u7684\u5f71\u54cd\u3002<\/p>\n\n\n\n

\uff085\uff09\u5206\u6790\u4e0d\u540c\u504f\u659c\u7a0b\u5ea6\u5bf9\u6bcf\u79cd\u591a\u6a21\u6001LLM\u63d0\u53d6\u7cbe\u5ea6\u7684\u5f71\u54cd<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5206\u6790\u6587\u6863\u504f\u8f6c\u89d2\u5ea6\u4e0e\u6a21\u578b\u8bc6\u522b\u7cbe\u786e\u5ea6\u7b49\u4fe1\u606f\u7684\u997c\u72b6\u56fe\u6211\u4eec\u53ef\u4ee5\u5f97\u51fa\uff1a<\/p>\n\n\n\n

1\uff09GPT-4-Turbo\u3001Anthropic Claude V3 Sonnet\u3001Llava:v1.6\u6587\u6863\u503e\u659c\u7684\u654f\u611f\u7a0b\u5ea6\u4e0d\u540c\u3002\u5176\u4e2dLlava:v1.6\u5bf9\u6587\u6863\u503e\u659c\u6700\u4e3a\u654f\u611f\uff0cGPT-4-Turbo\u5bf9\u6587\u6863\u503e\u659c\u7684\u9002\u5e94\u6027\u6700\u5927\u3002<\/p>\n\n\n\n

2\uff09\u5728\u6587\u6863\u503e\u659c\u8fbe\u5230\u4e00\u5b9a\u7a0b\u5ea6\u540e\uff0c\u5404\u4e2a\u6a21\u578b\u51fa\u73b0\u4e86\u4e0d\u540c\u7a0b\u5ea6\u7684\u5e7b\u89c9\u3002\u4ea7\u751f\u4e86\u9519\u8bef\u7684\u865a\u5047\u4fe1\u606f\u3002<\/p>\n\n\n\n

3\uff09\u5bf9\u4e8e\u503e\u659c\u6587\u6863\u5bf9\u5927\u6a21\u578b\u8bc6\u522b\u6548\u679c\u4ea7\u751f\u5f71\u54cd\u7684\u95ee\u9898\uff0c\u6587\u6863\u63d0\u51fa\u4e86\u4e24\u79cd\u53ef\u884c\u7684\u65b9\u6848\u4ee5\u6d88\u9664\u5176\u5f71\u54cd\u3002\u4e00\u662f\u901a\u8fc7\u5404\u79cd\u7b97\u6cd5\u5bf9\u6587\u6863\u8fdb\u884c\u9884\u5904\u7406\uff0c\u5c06\u6587\u6863\u503e\u659c\u63d0\u524d\u8bc6\u522b\u51fa\u6765\uff0c\u5e76\u5bf9\u5176\u8fdb\u884c\u7ea0\u6b63\uff1b\u800c\u662f\u5728\u8bad\u7ec3\u5927\u6a21\u578b\u65f6\uff0c\u5c06\u503e\u659c\u6587\u6863\u4e5f\u4f5c\u4e3a\u5176\u8bad\u7ec3\u6570\u636e\u7684\u4e00\u90e8\u5206\uff0c\u4f7f\u5927\u6a21\u578b\u80fd\u591f\u62e5\u6709\u5b8c\u5584\u7684\u8bc6\u522b\u5404\u7c7b\u503e\u659c\u6587\u6863\u7684\u80fd\u529b\uff0c\u4ee5\u5b9e\u73b0\u5bf9\u503e\u659c\u6587\u6863\u7684\u8bc6\u522b\u5904\u7406\u3002<\/p>\n\n\n\n

4.\u603b\u7ed3\u4e0e\u601d\u8003<\/h2>\n\n\n\n

1.\u672c\u6587\u7684\u6838\u5fc3\u601d\u8def\u662f\u5229\u7528\u591a\u6a21\u6001\u5927\u6a21\u578b\u6267\u884c\u7c7b\u4f3cOCR\u7684\u529f\u80fd\uff0c\u5229\u7528\u5927\u6a21\u578b\u5bf9\u4ee5\u56fe\u50cf\u4e3a\u8f7d\u4f53\u7684\u6587\u6863\u6570\u636e\u8fdb\u884c\u5904\u7406\uff0c\u4ee5\u8fdb\u884c\u51c6\u786e\u7684\u7ed3\u6784\u5316\u6570\u636e\u63d0\u53d6\u3002\u5206\u6790\u4e86\u6587\u6863\u503e\u659c\u56e0\u7d20\u5bf9\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc6\u522b\u6548\u679c\u7684\u8d1f\u9762\u5f71\u54cd\u4ee5\u53ca\u6d88\u9664\u5f71\u54cd\u7684\u65b9\u6cd5\u3002<\/p>\n\n\n\n

2.\u63d0\u4f9b\u4e86\u5229\u7528\u591a\u6a21\u6001\u5927\u6a21\u578b\u8fdb\u884c\u6570\u636e\u5904\u7406\u7684\u4e00\u79cd\u65b0\u7684\u7814\u7a76\u65b9\u6cd5\u548c\u7814\u7a76\u89c6\u89d2\u3002\u5728\u5229\u7528\u591a\u6a21\u6001\u5927\u6a21\u578b\u5904\u7406\u7f51\u9875\u4fe1\u606f\u6570\u636e\u548c\u5176\u4ed6\u5f62\u5f0f\u6570\u636e\u65f6\uff0c\u5e94\u6ce8\u610f\u7c7b\u4f3c\u201c\u6587\u6863\u503e\u659c\u201d\u56e0\u7d20\u5bf9\u6570\u636e\u63d0\u53d6\u5904\u7406\u6548\u679c\u7684\u5f71\u54cd\u3002<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

\u4f5c\u8005\uff1a Anjanava Biswas , Wrick Talukdar \u671f\u520a\uff1a Journal of Art […]<\/p>\n","protected":false},"author":62,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1316","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1316","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/users\/62"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=1316"}],"version-history":[{"count":4,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1316\/revisions"}],"predecessor-version":[{"id":1386,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1316\/revisions\/1386"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=1316"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=1316"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=1316"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}