{"id":1567,"date":"2024-11-13T10:51:35","date_gmt":"2024-11-13T02:51:35","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=1567"},"modified":"2024-11-13T10:51:35","modified_gmt":"2024-11-13T02:51:35","slug":"202411%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-contextual-document-embeddings","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202411%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-contextual-document-embeddings\/","title":{"rendered":"202411\u8bba\u6587\u7814\u8bfb-CONTEXTUAL DOCUMENT EMBEDDINGS"},"content":{"rendered":"\n

\u4f5c\u8005\uff1aJohn X. Morris\uff0cAlexander M. Rush<\/p>\n\n\n\n

\u5355\u4f4d\uff1aCornell University<\/p>\n\n\n\n

\u6765\u6e90\uff1aarXiv 2024<\/p>\n\n\n\n

\u4e00\u3001\u4e3b\u8981\u5185\u5bb9<\/p>\n\n\n\n

\u4e3b\u8981\u63a2\u8ba8\u5982\u4f55\u5728\u6587\u672c\u68c0\u7d22\u4efb\u52a1\u4e2d\u901a\u8fc7\u5f15\u5165\u4e0a\u4e0b\u6587\u4fe1\u606f\u6765\u63d0\u5347\u6587\u6863\u5d4c\u5165\u7684\u8868\u73b0\u3002\u4f5c\u8005\u8ba4\u4e3a\u4f20\u7edf\u53cc\u7f16\u7801\u5668\u67b6\u6784\u5728\u7f16\u7801\u65f6\u5ffd\u7565\u4e86\u6587\u6863\u95f4\u7684\u4e0a\u4e0b\u6587\u8054\u7cfb\uff0c\u5bfc\u81f4\u5728\u8de8\u57df\u6216\u7279\u5b9a\u57df\u68c0\u7d22\u4e2d\u7684\u6027\u80fd\u4e0d\u4f73\u3002\u4e3a\u6b64\uff0c\u63d0\u51fa\u4e86\u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60\uff08Contextual Contrastive Learning\uff09<\/strong> \u548c \u4e0a\u4e0b\u6587\u7f16\u7801\u67b6\u6784\uff08Contextual Encoder Architecture\uff09<\/strong> \u4e24\u79cd\u6280\u672f\u6765\u6539\u5584\u8fd9\u4e00\u95ee\u9898\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u8fd9\u79cd\u65b9\u6cd5\u5728\u591a\u4e2a\u57fa\u51c6\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u4f18\u4e8e\u65e0\u4e0a\u4e0b\u6587\u7684\u53cc\u7f16\u7801\u5668\uff0c\u5c24\u5176\u5728\u51fa\u57df\u68c0\u7d22\u4efb\u52a1\u4e2d\u5c55\u73b0\u4e86\u66f4\u5f3a\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n

    \n
  1. \u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60<\/strong>\uff1a\u901a\u8fc7\u751f\u6210\u90bb\u8fd1\u6587\u6863\u5e76\u5728\u6279\u6b21\u4e2d\u5f15\u5165\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u6587\u6863\uff0c\u4f7f\u6a21\u578b\u5728\u5bf9\u6bd4\u5b66\u4e60\u4e2d\u66f4\u597d\u5730\u5229\u7528\u8d1f\u6837\u672c\u3002<\/li>\n\n\n\n
  2. \u4e0a\u4e0b\u6587\u7f16\u7801\u67b6\u6784<\/strong>\uff1a\u901a\u8fc7\u4e24\u9636\u6bb5\u7684\u5d4c\u5165\u67b6\u6784\uff0c\u4e3b\u6587\u6863\u5d4c\u5165\u8fc7\u7a0b\u4e2d\u5f15\u5165\u90bb\u8fd1\u6587\u6863\u7684\u4fe1\u606f\uff0c\u4ee5\u751f\u6210\u66f4\u5177\u9886\u57df\u611f\u77e5\u7684\u5d4c\u5165\u3002<\/li>\n\n\n\n
  3. \u5b9e\u9a8c\u9a8c\u8bc1<\/strong>\uff1a\u5728\u4e0d\u540c\u4efb\u52a1\u548c\u6570\u636e\u96c6\u4e0a\u5bf9\u6a21\u578b\u8fdb\u884c\u4e86\u6d4b\u8bd5\uff0c\u5c24\u5176\u5728\u8de8\u57df\u4efb\u52a1\u548c\u7279\u5b9a\u9886\u57df\uff08\u5982\u533b\u7597\u548c\u91d1\u878d\uff09\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u663e\u8457\u63d0\u5347\u3002<\/li>\n<\/ol>\n\n\n\n

    \u4e8c\u3001\u65b9\u6cd5<\/p>\n\n\n\n

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

    \u4e00\u3001\u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60\uff08Contextual Contrastive Learning\uff09\uff1a<\/p>\n\n\n\n

    \u2022 \u672c\u6587\u91c7\u7528\u5feb\u901f\u805a\u7c7b\u7b97\u6cd5\u5c06\u67e5\u8be2-\u6587\u6863\u5bf9\u5206\u7ec4\uff0c\u4ee5\u751f\u6210\u90bb\u8fd1\u6587\u6863\u7684\u96c6\u5408\u3002<\/p>\n\n\n\n

    \u2022 \u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60\u7684\u635f\u5931\u51fd\u6570\uff1a\u5728\u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60\u4e2d\uff0c\u6a21\u578b\u5728\u6279\u6b21\u5185\u4ec5\u8003\u8651\u540c\u4e00\u4e0a\u4e0b\u6587\u6279\u6b21\u7684\u8d1f\u6837\u672c\uff0c\u800c\u4e0d\u4f9d\u8d56\u5168\u5c40\u8d1f\u6837\u672c<\/p>\n\n\n\n

    \u4e8c\u3001\u4e0a\u4e0b\u6587\u7f16\u7801\u67b6\u6784\uff08Contextual Encoder Architecture\uff09\uff1a
    \u2022 \u8bbe\u8ba1\u76ee\u6807\uff1a\u5728\u7f16\u7801\u6587\u6863\u65f6\uff0c\u5c06\u90bb\u8fd1\u6587\u6863\u7684\u4fe1\u606f\u6574\u5408\u5230\u6587\u6863\u7684\u5d4c\u5165\u751f\u6210\u4e2d\u3002
    \u2022 \u4e24\u9636\u6bb5\u5d4c\u5165<\/p>\n\n\n\n

    \u2022 \u7b2c\u4e00\u9636\u6bb5\uff1aM_1(d_1),M_1(d_2),\u2026,M_1(d_J)
    \u2022 \u7b2c\u4e8c\u9636\u6bb5\uff1a\u03d5(d\u2032;D)=M_2(M_1(d_1),\u2026,M_1(d_J),E(d\u2032_1),\u2026,E(d\u2032_T))<\/p>\n\n\n\n

    \u4e09\u3001\u5b9e\u9a8c<\/p>\n\n\n\n

    \u2022 \u6a21\u578b\u8bad\u7ec3\uff1a<\/strong>
    \u2022 \u5c0f\u89c4\u6a21\u5b9e\u9a8c\uff1a\u5728\u5c0f\u89c4\u6a21\u7684BEIR\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u591a\u79cd\u6279\u6b21\u548c\u805a\u7c7b\u89c4\u6a21\u7684\u5b9e\u9a8c\uff0c\u4ee5\u9a8c\u8bc1\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u4e0a\u4e0b\u6587\u7f16\u7801\u7684\u6709\u6548\u6027\u3002
    \u2022 \u5927\u89c4\u6a21\u5b9e\u9a8c\uff1a\u5728\u5b8c\u6574\u7684MTEB\u57fa\u51c6\u4e0a\u8fdb\u884c\u5b9e\u9a8c\uff0c\u4ee5\u8bc4\u4f30\u8be5\u65b9\u6cd5\u5728\u66f4\u5927\u89c4\u6a21\u68c0\u7d22\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u3002MTEB\u5305\u542b\u4e86\u6587\u672c\u68c0\u7d22\u3001\u5206\u7c7b\u3001\u805a\u7c7b\u7b49\u591a\u79cd\u4efb\u52a1\uff0c\u8986\u76d6\u4e86\u533b\u7597\u3001\u91d1\u878d\u7b49\u591a\u4e2a\u9886\u57df\u3002
    \u2022 \u8bad\u7ec3\u6570\u636e\u4e0e\u6307\u6807\uff1a<\/strong>
    \u2022 \u8bad\u7ec3\u6570\u636e\uff1a\u4f7f\u7528200M\u7684\u5f31\u76d1\u7763\u6570\u636e\uff0c\u540c\u65f6\u57281.8M\u4eba\u5de5\u6807\u6ce8\u7684\u67e5\u8be2-\u6587\u6863\u5bf9\u4e0a\u8fdb\u884c\u6709\u76d1\u7763\u5fae\u8c03\u3002
    \u2022 \u8bc4\u4f30\u6307\u6807\uff1aNDCG@10<\/p>\n\n\n\n

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

    \u56db\u3001\u603b\u7ed3<\/p>\n\n\n\n

    \u8fd9\u7bc7\u8bba\u6587\u63a2\u8ba8\u5982\u4f55\u5728\u6587\u672c\u68c0\u7d22\u4efb\u52a1\u4e2d\u901a\u8fc7\u5f15\u5165\u4e0a\u4e0b\u6587\u4fe1\u606f\u6765\u63d0\u5347\u6587\u6863\u5d4c\u5165\u7684\u8868\u73b0\u3002\u4f20\u7edf\u53cc\u7f16\u7801\u5668\u67b6\u6784\u5728\u7f16\u7801\u65f6\u5ffd\u7565\u4e86\u6587\u6863\u95f4\u7684\u4e0a\u4e0b\u6587\u8054\u7cfb\uff0c\u5bfc\u81f4\u5728\u8de8\u57df\u6216\u7279\u5b9a\u57df\u68c0\u7d22\u4e2d\u7684\u6027\u80fd\u4e0d\u4f73\u3002
    \u4e3a\u6b64\uff0c\u63d0\u51fa\u4e86\u4e0a\u4e0b\u6587\u5bf9\u6bd4\u5b66\u4e60\uff08Contextual Contrastive Learning\uff09 \u548c \u4e0a\u4e0b\u6587\u7f16\u7801\u67b6\u6784\uff08Contextual Encoder Architecture\uff09\u3002\u8fd9\u79cd\u65b9\u6cd5\u5728\u591a\u4e2a\u57fa\u51c6\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u4f18\u4e8e\u65e0\u4e0a\u4e0b\u6587\u7684\u53cc\u7f16\u7801\u5668\uff0c\u5c24\u5176\u5728\u51fa\u57df\u68c0\u7d22\u4efb\u52a1\u4e2d\u5c55\u73b0\u4e86\u66f4\u5f3a\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n

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

    \u4f5c\u8005\uff1aJohn X. Morris\uff0cAlexander M. Rush \u5355\u4f4d\uff1aCornell Universi […]<\/p>\n","protected":false},"author":9,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1567","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1567","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\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=1567"}],"version-history":[{"count":4,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1567\/revisions"}],"predecessor-version":[{"id":1574,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/1567\/revisions\/1574"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=1567"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=1567"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=1567"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}