{"id":17854,"date":"2024-01-22T18:06:30","date_gmt":"2024-01-22T10:06:30","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=17854"},"modified":"2024-01-22T18:07:42","modified_gmt":"2024-01-22T10:07:42","slug":"%e9%92%88%e5%af%b9%e7%89%b9%e5%ae%9a%e9%a2%86%e5%9f%9f%e8%be%83%e5%b0%8f%e7%9a%84%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e6%98%af%e5%90%a6%e4%b8%8e%e8%be%83%e5%a4%a7%e7%9a%84%e6%a8%a1%e5%9e%8b%e5%90%8c","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2024\/01\/22\/17854\/","title":{"rendered":"\u9488\u5bf9\u7279\u5b9a\u9886\u57df\u8f83\u5c0f\u7684\u8bed\u8a00\u6a21\u578b\u662f\u5426\u4e0e\u8f83\u5927\u7684\u6a21\u578b\u540c\u6837\u6709\u6548?"},"content":{"rendered":"
\u7ecf\u8fc72023\u5e74\u7684\u53d1\u5c55\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u5c55\u793a\u51fa\u4e86\u975e\u5e38\u5927\u7684\u6f5c\u529b\uff0c\u8bad\u7ec3\u8d8a\u6765\u8d8a\u5927\u7684\u6a21\u578b\u6210\u4e3a\u6709\u6548\u6027\u8bc4\u4f30\u7684\u4e00\u4e2a\u5173\u952e\u6307\u6807\uff0c\u8bba\u6587\u300aA Comparative Analysis of Fine-Tuned LLMs and Few-Shot Learning of LLMs for Financial Sentiment Analysis\u300b\u4e2d\uff0c\u4f5c\u8005\u5168\u9762\u5206\u6790\u4e86\u5fae\u8c03\u5927\u8bed\u8a00\u6a21\u578b(llm)\u53ca\u5176\u5728\u91d1\u878d\u60c5\u7eea\u5206\u6790\u4e2d\u7684\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u7684\u80fd<\/div>\n
\n
\n

\u7ecf\u8fc72023\u5e74\u7684\u53d1\u5c55\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u5c55\u793a\u51fa\u4e86\u975e\u5e38\u5927\u7684\u6f5c\u529b\uff0c\u8bad\u7ec3\u8d8a\u6765\u8d8a\u5927\u7684\u6a21\u578b\u6210\u4e3a\u6709\u6548\u6027\u8bc4\u4f30\u7684\u4e00\u4e2a\u5173\u952e\u6307\u6807\uff0c\u8bba\u6587\u300aA Comparative Analysis of Fine-Tuned LLMs and Few-Shot Learning of LLMs for Financial Sentiment Analysis\u300b\u4e2d\uff0c\u4f5c\u8005\u5168\u9762\u5206\u6790\u4e86\u5fae\u8c03\u5927\u8bed\u8a00\u6a21\u578b(llm)\u53ca\u5176\u5728\u91d1\u878d\u60c5\u7eea\u5206\u6790\u4e2d\u7684\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u7684\u80fd\u529b\u3002<\/p>\n

\"\"<\/p>\n

\u4f5c\u8005\u901a\u8fc7\u4f7f\u7528\u4e24\u79cd\u65b9\u6cd5\u6765\u63a2\u7d22\u91d1\u878d\u60c5\u7eea\u5206\u6790\u80cc\u666f\u4e0b\u7684\u6f5c\u529b\u548c\u9002\u7528\u6027:<\/p>\n

    \n
  • \u5728\u7279\u5b9a\u7684\u9886\u57df\uff08\u91d1\u878d\u9886\u57df\uff09\u7684\u6570\u636e\u96c6\u4e0a\uff0c\u4f7f\u7528\u5c0f\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff0c\u4f5c\u8005\u6d4b\u8bd5\u4e86250M\u52303B\u53c2\u6570\u5404\u79cd\u6a21\u578b<\/li>\n
  • \u4ee5gpt-3.5 turbo\u4e3a\u91cd\u70b9\u7684\u60c5\u5883\u5b66\u4e60<\/li>\n<\/ul>\n

    \u4f5c\u8005\u8fd8\u5c06\u7ed3\u679c\u4e0eSOTA(\u6700\u5148\u8fdb\u7684)\u6a21\u578b\u8fdb\u884c\u6bd4\u8f83\u4ee5\u8bc4\u4f30\u5176\u6027\u80fd\uff0c\u6211\u4eec\u770b\u770b\u5c0f\u6a21\u578b\u662f\u5426\u8fd8\u540c\u6837\u6709\u6548\u3002<\/p>\n

    \u8bba\u6587\u8bc1\u660e\u4e86\u4ee5\u4e0b\u89c2\u70b9:<\/p>\n

      \n
    • \u5fae\u8c03\u8f83\u5c0f\u7684llm\u53ef\u4ee5\u8fbe\u5230\u4e0eSOTA\u5fae\u8c03llm\u76f8\u5f53\u7684\u6027\u80fd\u3002<\/li>\n
    • \u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u5b66\u4e60\u7684\u7684\u6027\u80fd\u4e0e\u7ecf\u8fc7\u5fae\u8c03\u7684\u5c0f\u578bllm\u76f8\u5f53\u3002<\/li>\n
    • \u589e\u52a0\u4e0a\u4e0b\u6587\u5b66\u4e60\u4e2d\u7684\u6837\u672c\u6570\u91cf\u5e76\u4e0d\u4e00\u5b9a\u4f1a\u63d0\u9ad8\u60c5\u611f\u5206\u6790\u4efb\u52a1\u7684\u6027\u80fd\u3002<\/li>\n
    • \u5fae\u8c03\u8f83\u5c0f\u7684llm\u4f1a\u964d\u4f4e\u6210\u672c\u548c\u63d0\u9ad8\u8ba1\u7b97\u6548\u7387\u3002<\/li>\n<\/ul>\n

      \u4f5c\u8005\u4e13\u6ce8\u4e8e\u4f7f\u7528QLoRa (Quantized low – rank – adaptive)\u673a\u5236\u5bf9FLAN-T5\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u3002\u4f7f\u7528\u8d22\u52a1\u7279\u5b9a\u6570\u636e\u96c6\uff0c\u7814\u7a76\u4e863\u79cd\u5c3a\u5bf8:Flan-T5 base (250M)\uff0c Flan-T5 large (780M)\u548cFlan-T5-xl (3B\u53c2\u6570)\u3002<\/p>\n

      \u8bba\u6587\u6982\u8ff0<\/h4>\n

      \u8bba\u6587\u9996\u5148\u603b\u7ed3\u4e86\u7279\u5b9a\u4e8e\u91d1\u878d\u9886\u57df\u7684SOTA\u6a21\u578b:<\/p>\n

        \n
      • FinBERT:\u4f7f\u7528\u603b\u8ba14.9B Token\u7ec4\u7684\u91d1\u878d\u901a\u4fe1\u8bed\u6599\u5e93\u8fdb\u884c\u5fae\u8c03\u7684BERT\u3002<\/li>\n
      • bloomberg ggpt:\u8fd9\u662f\u4e00\u4e2a\u5305\u542b50B\u4e2a\u53c2\u6570\u7684\u95ed\u6e90\u6a21\u578b\uff0c\u4e13\u95e8\u9488\u5bf9\u5404\u79cd\u91d1\u878d\u6570\u636e\u8fdb\u884c\u8bad\u7ec3\u3002\u5b83\u5728\u60c5\u611f\u5206\u6790\u4e2d\u8868\u73b0\u51fa\u826f\u597d\u7684\u6027\u80fd\u3002<\/li>\n
      • \u4f7f\u7528LLama-7B\u5bf9FinGPT\u8fdb\u884c\u5fae\u8c03\u3002\u8be5\u6a21\u578b\u4f7f\u7528\u66f4\u5c11\u7684\u8ba1\u7b97\u8d44\u6e90\u5b9e\u73b0\u4e86\u4e0ebloomberg ggpt\u76f8\u5f53\u7684\u6027\u80fd\u3002<\/li>\n
      • ChatGPT\u8fd9\u6837\u7684llm\u4e5f\u53ef\u4ee5\u4f7f\u7528\u96f6\u6837\u672c\u5b66\u4e60\u3002\u4f46\u662f\u4ed6\u4eec\u5728\u5c11\u6837\u672c\u5b66\u4e60\u4e2d\u8868\u73b0\u5e76\u4e0d\u7406\u60f3<\/li>\n<\/ul>\n

        \u4f5c\u8005\u4f7f\u7528\u4e86\u4ee5\u4e0b\u6a21\u578b\uff1a<\/p>\n

        1\u3001\u6ca1\u6709\u8fdb\u884c\u4efb\u4f55\u5fae\u8c03:Flan-T5 base (250M)\uff0c Flan-T5 large (780M)\uff0c Flan-T5-xl (3B\u53c2\u6570)\uff0cChatGPT (gpt-3.5 turbo)\u3002\u76ee\u6807\u662f\u7814\u7a76\u6a21\u578b\u7684\u5927\u5c0f\u5bf9\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u5b66\u4e60\u7684\u5f71\u54cd\u3002<\/p>\n

        2\u3001\u5fae\u8c03llm:\u5177\u67093\u4e2a\u5c3a\u5bf8\u7684\u76f8\u540c\u578b\u53f7\u7684Flan-T5\u5df2\u7ecf\u8fdb\u884c\u4e86\u5fae\u8c03\u3002<\/p>\n

        \u6570\u636e\u96c6<\/h4>\n

        \u4f7f\u7528\u4e86Twitter\u8d22\u7ecf\u65b0\u95fb(Twitter Train)\uff0c\u5305\u62ec\u4e0e\u91d1\u878d\u4e3b\u9898\u76f8\u5173\u7684\u63a8\u6587\uff0c\u53ef\u901a\u8fc7HuggingFace\u8bbf\u95ee\u3002\u5b83\u5305\u542b9540\u4e2a\u6837\u672c\u3002<\/p>\n

        TFSN: 2390\u4e2a\u5e26\u6709\u6ce8\u91ca\u7684\u8d22\u7ecf\u76f8\u5173\u63a8\u6587\u8bed\u6599\u5e93\u6837\u672c\u3002<\/p>\n

        FPB: 4845\u7bc7\u91d1\u878d\u65b0\u95fb\u6587\u7ae0\u6837\u672c\uff0c\u753116\u4f4d\u9886\u57df\u4e13\u5bb6\u6ce8\u91ca\u3002<\/p>\n

        GPU\u8d44\u6e90<\/h4>\n

        \u4e3a\u4e86\u5bf93\u4e2a\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff0c\u4f5c\u8005\u4f7f\u7528\u4e86A100 GPU\uff0c\u6bcf\u4e2a\u6a21\u578b\u7684\u603b\u8bad\u7ec3\u65f6\u95f4\u5982\u4e0b:\u57fa\u672c\u6a21\u578b28\u5206\u949f\uff0c\u5927\u6a21\u578b54\u5206\u949f\uff0cXL\u6a21\u578b65\u5206\u949f\uff0c\u6240\u4ee5\u8bf4\u8fd9\u4e2a\u5fae\u8c03\u662f\u975e\u5e38\u8282\u7701\u8d44\u6e90\u7684\u3002<\/p>\n

        \u5fae\u8c03\u5c0f\u578bLLMs<\/h4>\n

        \u7ed3\u679c\u663e\u793a\u4e86\u7ecf\u8fc7\u5fae\u8c03\u7684\u5c0f\u578bllm\u4f18\u4e8e\u5927\u578bllm\u7684\u6027\u80fd:<\/p>\n

        \"\"<\/p>\n

        \u6240\u6709Fine-tuned-FLAN-T5\u7684\u6027\u80fd\u90fd\u4f18\u4e8eFinBERT\uff1bLarge (780M)\u548cXL(3B) fine – tuning – flan – t5\u6027\u80fd\u4f18\u4e8edirective – lama- 7b\uff1b\u5728TFSN\u6570\u636e\u96c6\u4e2d\uff0c\u5373\u4f7f\u662f\u57fa\u7840(250M)\u5fae\u8c03\u7684flan – t5\u4e5f\u6bd4\u4f7f\u7528ChatGPT (gpt-3.5 turbo)\u7684k-shot\u4e0a\u4e0b\u6587\u5b66\u4e60\u8868\u73b0\u66f4\u597d\u3002<\/p>\n

        \u5c11\u6837\u672c\u5b66\u4e60<\/h4>\n

        \u4ee5\u4e0b\u662f0 -shot\u548ck-shot\u5b66\u4e60\u7684\u7ed3\u679c(k= 1,5\u548c10):<\/p>\n

        \"\"<\/p>\n

        \u5728TFSN\u6570\u636e\u96c6\u4e0a\uff0c\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u5b66\u4e60\u7684\u8868\u73b0\u660e\u663e\u4f4e\u4e8e\u6240\u6709\u5176\u4ed6\u5fae\u8c03\u6a21\u578b\u3002(\u9664\u4e86XL\uff0c\u8868\u73b0\u6bd4FinBert\u7a0d\u597d)<\/p>\n

        \u4f46\u662f\u5728FPB\u6570\u636e\u96c6\u4e2d\uff0c\u4e0eChatGPT\u76f8\u6bd4\uff0cLarge\u548cXL Flan-T5\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u5b66\u4e60\u8868\u73b0\u51fa\u8f83\u5f3a\u7684\u6027\u80fd\u3002<\/p>\n

        \u6837\u672c\u7684\u589e\u52a0\u4f7f\u5f97\u57fa\u672c\u578bFlan-T5\u7684\u6027\u80fd\u7565\u6709\u63d0\u5347\u3002\u4f46\u5728Large\u548cXL fall – t5\u4e2d\uff0c\u8fd9\u5bfc\u81f4\u7cbe\u5ea6\u4e0b\u964d\u3002\u8fd9\u662f\u56e0\u4e3a\u5197\u957f\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u53ef\u80fd\u4f1a\u5bfc\u81f4LLM\u8bef\u5165\u6b67\u9014\u3002<\/p>\n

        \u6240\u4ee5\u4f5c\u8005\u5efa\u8bae\uff0c\u5f53k-shot\u589e\u52a0\u65f6\u53ef\u4ee5\u4f7f\u7528\u8bed\u4e49\u76f8\u4f3c\u68c0\u7d22\u6216\u601d\u7ef4\u94fe(CoT)\u6216\u7ebf\u7d22\u548c\u63a8\u7406\u63d0\u793a(CARP)\u65b9\u6cd5\u6765\u89e3\u51b3\u6027\u80fd\u4e0b\u964d\u7684\u95ee\u9898\u3002<\/p>\n

        \u603b\u7ed3<\/h4>\n

        \u53ef\u4ee5\u770b\u5230\uff0c\u9488\u5bf9\u7279\u5b9a\u7684\u9886\u57df\uff0c\u5fae\u8c03\u5c0f\u6a21\u578b\u8fd8\u662f\u80fd\u8fc7\u591f\u5f97\u5230\u5f88\u597d\u7684\u6548\u679c\uff0c\u8fd9\u5728\u5bf9\u4e8e\u6211\u4eec\u5b9e\u9645\u5e94\u7528\u662f\u662f\u975e\u5e38\u6709\u5e2e\u52a9\u7684\uff0c\u4e0d\u4ec5\u53ef\u4ee5\u8282\u7701\u6210\u672c\uff0c\u8fd8\u53ef\u4ee5\u8282\u7701\u6211\u4eec\u7684\u8bad\u7ec3\u65f6\u95f4\uff0c\u53ef\u4ee5\u8ba9\u6211\u4eec\u8fdb\u884c\u5feb\u901f\u7684\u7248\u672c\u8fed\u4ee3\u3002<\/p>\n

        \u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2312.08725.pdf<\/p>\n

        \u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

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

        \u7ecf\u8fc72023\u5e74\u7684\u53d1\u5c55\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u5c55\u793a\u51fa\u4e86\u975e\u5e38\u5927\u7684\u6f5c\u529b\uff0c\u8bad\u7ec3\u8d8a\u6765\u8d8a\u5927\u7684\u6a21\u578b\u6210\u4e3a\u6709\u6548\u6027\u8bc4\u4f30\u7684\u4e00\u4e2a\u5173\u952e\u6307\u6807\uff0c\u8bba\u6587\u300aA […]<\/p>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

        <\/div>\n","protected":false},"author":56,"featured_media":17855,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,23,80],"tags":[113,968],"class_list":["post-17854","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-23","category-80","tag-113","tag-968"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/17854","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=17854"}],"version-history":[{"count":3,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/17854\/revisions"}],"predecessor-version":[{"id":17863,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/17854\/revisions\/17863"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/17855"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=17854"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=17854"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=17854"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}