
{"id":10505,"date":"2026-07-28T19:41:57","date_gmt":"2026-07-28T11:41:57","guid":{"rendered":"https:\/\/infernews.com\/blog\/acl-2026-leveraging-external-knowledge-for-historical-document-restoration-via-r\/"},"modified":"2026-07-28T21:27:11","modified_gmt":"2026-07-28T13:27:11","slug":"acl-2026-leveraging-external-knowledge-for-historical-document-restoration-via-r","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/acl-2026-leveraging-external-knowledge-for-historical-document-restoration-via-r\/","title":{"rendered":"ARI \u7528 RAG \u4fee\u5fa9\u97d3\u570b\u671d\u9bae\u53e4\u7c4d\u6b98\u5b57"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/07\/pasted-7d787893f18b.jpg\" alt=\"Method Figure\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6700\u503c\u5f97\u7559\u610f\u7684\uff0c\u4e0d\u662f\u6a21\u578b\u628a\u7f3a\u5b57\u88dc\u56de\u4f86\u672c\u8eab\uff0c\u800c\u662f\u5b83\u5c08\u9580\u8655\u7406\u53e4\u7c4d\u4fee\u5fa9\u6700\u68d8\u624b\u7684\u4e00\u985e\u5167\u5bb9\uff1a\u4eba\u540d\u3001\u5730\u540d\u7b49 Named Entities\u3002ARI \u5c6c\u65bc\u4e00\u500b\u7d50\u5408 Retrieval-Augmented Generation\uff08RAG\uff09\u7684\u6587\u737b\u4fee\u5fa9\u6846\u67b6\uff0c\u91dd\u5c0d\u671d\u9bae\u738b\u671d\u5be6\u9304\u8207\u627f\u653f\u9662\u65e5\u8a18\u9019\u985e\u97d3\u6587\u6f22\u5b57\u53f2\u6599\uff0c\u88dc\u8db3\u53ea\u9760\u5c40\u90e8\u8a9e\u5883\u6642\u7d93\u5e38\u5931\u6e96\u7684\u7f3a\u53e3\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709\u505a\u6cd5\u591a\u6578\u4f9d\u8cf4 masked language modeling\uff0c\u64c5\u9577\u6839\u64da\u524d\u5f8c\u6587\u731c\u6e2c\u4e00\u822c\u5b57\u8a5e\uff0c\u4f46\u4e00\u9047\u5230\u9700\u8981\u5916\u90e8\u53f2\u5be6\u652f\u6301\u7684\u5c08\u540d\u5c31\u5bb9\u6613\u5931\u624b\u3002ARI \u7684\u53d6\u5411\u5f88\u6e05\u695a\uff1a\u5148\u7528 BM25 \u5f9e\u6b77\u53f2\u8a9e\u6599\u627e\u51fa\u524d 20 \u4efd\u76f8\u95dc\u6587\u672c\uff0c\u518d\u4ee5\u5b57\u4e32\u76f8\u4f3c\u5ea6 0.8 \u904e\u6ffe\u91cd\u8907\u5167\u5bb9\uff0c\u5c07\u9019\u4e9b\u5916\u90e8\u8b49\u64da\u4ea4\u7d66\u6a21\u578b\u4e00\u4f75\u751f\u6210\uff0c\u4fee\u6b63\u901a\u7528 LLM \u5bb9\u6613\u51fa\u73fe\u7684\u5e7b\u89ba\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u90e8\u5206\u4e0d\u662f\u5f9e\u96f6\u958b\u59cb\uff0c\u800c\u662f\u5efa\u57fa\u65bc Qwen3 32B \u8207 Qwen3 8B \u5fae\u8abf\u6210 ARI-32B \u548c ARI-8B\uff0c\u4e26\u52a0\u5165 25% named entity-prioritized masking \u8a13\u7df4\u7b56\u7565\uff0c\u628a\u5b78\u7fd2\u91cd\u9ede\u653e\u5728\u77e5\u8b58\u5bc6\u96c6\u7247\u6bb5\u3002\u8ad6\u6587\u4ea6\u6307\u51fa\uff0c\u5c0d\u6f22\u5b57\u6750\u6599\u800c\u8a00\uff0c\u8a5e\u5f59\u5c64\u9762\u7684 BM25 \u6aa2\u7d22\u6bd4 embedding-based retrieval \u66f4\u6709\u6548\uff0c\u9019\u4e00\u9ede\u9817\u6709\u8aaa\u670d\u529b\uff0c\u56e0\u70ba\u8868\u610f\u6587\u5b57\u7684\u5b57\u5f62\u8207\u5b57\u8a5e\u5c0d\u61c9\u95dc\u4fc2\u672c\u8eab\u5c31\u5f71\u97ff\u6aa2\u7d22\u6548\u679c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u9069\u5408\u6b77\u53f2\u6587\u737b\u6574\u7406\u3001\u6578\u4f4d\u4eba\u6587\u7814\u7a76\u8207\u53e4\u7c4d\u6821\u52d8\u5718\u968a\u53c3\u8003<\/li>\n<li>\u4e3b\u8981\u5f37\u9805\u5728\u65bc\u4fee\u5fa9\u9700\u8981\u5916\u90e8\u77e5\u8b58\u652f\u6490\u7684 Named Entities<\/li>\n<li>ARI-32B \u8207 ARI-8B \u540c\u6b65\u63d0\u4f9b\uff0c\u524d\u8005\u8ffd\u6c42\u8868\u73fe\uff0c\u5f8c\u8005\u8f03\u91cd\u8996\u904b\u7b97\u6210\u672c<\/li>\n<li>\u8ad6\u6587\u7d50\u679c\u986f\u793a\uff0c\u5b83\u5728 named entity \u8207\u96a8\u6a5f\u906e\u7f69\u5b57\u5143\u4fee\u5fa9\u90fd\u52dd\u904e\u591a\u500b\u57fa\u7dda\u8207\u901a\u7528\u6a21\u578b<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u628a\u5b83\u8996\u70ba\u4e00\u500b\u5df2\u6709\u516c\u958b\u6a21\u578b\u8207\u65b9\u6cd5\u8aaa\u660e\u7684\u7814\u7a76\u9805\u76ee\u3002\u5c0d\u9700\u8981\u5148\u9a57\u8b49\u6548\u679c\u7684\u4eba\u4f86\u8aaa\uff0c\u73fe\u968e\u6bb5\u8f03\u5408\u7406\u7684\u8def\u7dda\u6703\u662f\u5148\u67e5\u770b\u8ad6\u6587\u8a2d\u5b9a\u8207\u6a21\u578b\u9801\u9762\uff0c\u518d\u5224\u65b7\u662f\u5426\u8db3\u4ee5\u63a5\u5165\u81ea\u5df1\u7684\u53e4\u7c4d\u4fee\u5fa9\u5de5\u4f5c\u6d41\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/huggingface.co\/collections\/DAMI-Lab\/ari-archive-restoration-intelligence\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/EvelynKimm\/ARI\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2607.21936\" rel=\"noopener noreferrer\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9047\u5230\u4eba\u540d\u3001\u5730\u540d\u6b98\u7f3a\u7684\u53e4\u7c4d\uff0c\u55ae\u9760\u4e0a\u4e0b\u6587\u5f80\u5f80\u6703\u4f30\u932f\u3002ARI\u628a\u6aa2\u7d22\u5230\u7684\u53f2\u6599\u4e00\u4f75\u4ea4\u7540\u6a21\u578b\u5224\u65b7\uff0c\u660e\u986f\u66f4\u9069\u5408\u505a\u6b77\u53f2\u6587\u737b\u4fee\u5fa9\u3002<\/p>\n","protected":false},"author":8,"featured_media":10504,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,185,131,109,76,128,199],"tags":[],"class_list":["post-10505","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-embedding","category-rag","category-76","category-128","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10505","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=10505"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10505\/revisions"}],"predecessor-version":[{"id":10513,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10505\/revisions\/10513"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10504"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10505"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10505"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10505"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}