
{"id":9657,"date":"2026-06-26T15:13:17","date_gmt":"2026-06-26T07:13:17","guid":{"rendered":"https:\/\/infernews.com\/blog\/remmdbench-agent\/"},"modified":"2026-06-26T15:13:17","modified_gmt":"2026-06-26T07:13:17","slug":"remmdbench-agent","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/remmdbench-agent\/","title":{"rendered":"ReMMDBench-Agent \u9a57\u8b49\u591a\u6a21\u614b\u5047\u8cc7\u8a0a"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-2faee779f2ef.jpg\" alt=\"Repository image for DANG-ai\/ReMMDBench-Agent\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u958b\u767c\u5718\u968a\u4f86\u81ea\u4e0a\u6d77\u4ea4\u901a\u5927\u5b78\u3001\u4e0a\u6d77\u4eba\u5de5\u667a\u6167\u5be6\u9a57\u5ba4\u3001\u6e05\u83ef\u5927\u5b78\u3001\u4e2d\u5357\u5927\u5b78\uff0c\u4ee5\u53ca\u4e2d\u570b\u96fb\u5b50\u79d1\u6280\u96c6\u5718\u7b2c\u5341\u4e94\u7814\u7a76\u6240\uff0c\u6838\u5fc3\u4f5c\u8005\u628a ReMMDBench \u540c ReMMD-Agent \u4e00\u8d77\u516c\u958b\uff0c\u65b9\u5411\u5f88\u660e\u78ba\uff1a\u7528\u8f03\u63a5\u8fd1\u771f\u5be6\u7db2\u7d61\u5e16\u6587\u7684\u65b9\u5f0f\uff0c\u6aa2\u67e5\u5716\u6587\u6df7\u5408\u5167\u5bb9\u4e2d\u7684 misinformation\u3002\u9019\u500b GitHub \u9805\u76ee\u5c6c\u65bc<strong>\u7814\u7a76\u539f\u578b\u52a0\u8a55\u6e2c\u4ee3\u78bc\u96c6\u5408<\/strong>\uff0c\u4e3b\u8981\u7528\u4f86\u91cd\u73fe\u4e09\u500b multimodal misinformation detection agent \u7cfb\u7d71\u5728 ReMMDBench \u4e0a\u7684\u7d50\u679c\uff0c\u4e26\u6bd4\u8f03\u5b83\u5011\u600e\u6a23\u505a\u5224\u65b7\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709\u505a\u6cd5\u5e38\u628a\u591a\u6a21\u614b\u5047\u8cc7\u8a0a\u6aa2\u6e2c\u6536\u7a84\u6210\u55ae\u5716\u3001\u4e8c\u5206\u985e\uff0c\u6216\u8005\u4e00\u6b21\u904e\u628a\u6574\u6bb5\u6587\u5b57\u8207\u5716\u7247\u4e1f\u7d66\u6a21\u578b\u5224\u65b7\uff1b\u4f5c\u8005\u8a8d\u70ba\u9019\u7a2e fixed-pass \u5224\u65b7\u65b9\u5f0f\u96e3\u4ee5\u8655\u7406\u9577\u6558\u4e8b\u3001\u591a\u5f35\u5716\u7247\u3001\u8de8\u8a9e\u8a00\u8207\u90e8\u5206\u771f\u5be6\u5167\u5bb9\u3002\u9019\u500b\u9805\u76ee\u56e0\u6b64\u63d0\u51fa\u4e00\u5957\u4ee5 ReMMDBench \u70ba\u6838\u5fc3\u7684 agentic \u9a57\u8b49\u8def\u7dda\uff1aBaseline 1 \u662f 3-stage MMD-Agent\uff0cBaseline 2 \u662f MCTS-based 5-verdict + 8-taxonomy agent\uff0c\u800c\u4e3b\u7cfb\u7d71 ReMMD-Agent \u5247\u7528 atomic decomposition\u3001RAG\uff08Retrieval-Augmented Generation\uff09\u8207 multi-expert judge\uff0c\u628a\u7d50\u8ad6\u5efa\u7acb\u5728\u53ef\u8ffd\u8e64\u7684\u8b49\u64da\u72c0\u614b\u4e0a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8ddf\u540c\u985e\u65b9\u6cd5\u76f8\u6bd4\uff0cReMMD-Agent \u7684\u53d6\u5411\u4e0d\u662f\u53ea\u8ffd\u6c42\u4e00\u6b21\u7b54\u4e2d\uff0c\u800c\u662f\u5148\u628a\u5e16\u6587\u62c6\u6210 atomic claims\u3001image observations\u3001text-image bindings\uff0c\u518d\u6aa2\u7d22 multimodal evidence\uff0c\u4e4b\u5f8c\u91cd\u7528 persistent memory\uff0c\u6e1b\u5c11\u91cd\u8907\u5de5\u5177\u547c\u53eb\u3002\u9019\u7a2e\u8a2d\u8a08\u7684\u53d6\u6368\u5f88\u6e05\u695a\uff1a\u6d41\u7a0b\u66f4\u9577\u3001\u914d\u7f6e\u66f4\u591a\uff0c\u4f46\u63db\u4f86\u8f03\u597d\u7684\u53ef\u89e3\u91cb\u6027\uff0c\u4e5f\u66f4\u9069\u5408\u8655\u7406 five-way L1 veracity labels\u30018 \u500b L2 distortion labels\uff0c\u4ee5\u53ca multilingual multi-image \u5834\u666f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b89\u88dd\u8207\u6e2c\u8a66\u601d\u8def\u4e5f\u76f8\u7576\u5177\u9ad4\u3002\u4e09\u500b\u5b50\u9805\u76ee\u5404\u81ea\u6709 <code>requirements.txt<\/code>\u3001\u8a2d\u5b9a\u6a94\u8207\u555f\u52d5\u8173\u672c\uff1b\u8981\u5148\u628a\u8cc7\u6599\u6839\u76ee\u9304\u6307\u5411 ReMMDBench\uff0c\u518d\u5728 <code>.yaml<\/code> \u6216 <code>.env<\/code> \u5167\u586b\u5165\u6a21\u578b\u7aef\u9ede\u8207\u91d1\u9470\u4f54\u4f4d\u5167\u5bb9\uff0c\u4e4b\u5f8c\u53ef\u5148\u7528 <code>mmd-agent\/test_qwen.py<\/code> \u9019\u985e\u5065\u5eb7\u6aa2\u67e5\u78ba\u8a8d\u5f8c\u7aef\u53ef\u56de\u61c9\uff0c\u518d\u8dd1\u5404\u81ea\u7684 evaluation scripts\u3002\u5009\u5eab\u5df2\u9644\u4e0a Qwen-family \u5f8c\u7aef\u7684\u4fdd\u5b58\u7d50\u679c\u8207 artifacts\uff0c\u5305\u542b Qwen 4B\u30019B\u300127B\uff0c\u4ea6\u660e\u78ba\u6a19\u793a temperature = 0.0\u3001LLM caching \u8207\u9810\u5efa RAG index\uff0c\u65b9\u4fbf\u91cd\u73fe headline numbers\uff0c\u800c\u4e0d\u5fc5\u7531\u96f6\u958b\u59cb\u5efa\u7acb\u6574\u5957\u6d41\u7a0b\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u4e3b\u7cfb\u7d71<\/strong>\uff1aReMMD-Agent\uff0c\u6838\u5fc3\u7d50\u69cb\u662f atomic decomposition + RAG + multi-expert judge<\/li>\n<li><strong>\u5c0d\u7167\u7cfb\u7d71<\/strong>\uff1a3-stage MMD-Agent \u8207 MCTS-based t2-agent\uff0c\u65b9\u4fbf\u770b\u4e0d\u540c agent \u8a2d\u8a08\u7684\u53d6\u6368<\/li>\n<li><strong>\u8cc7\u6599\u8207\u6a19\u8a3b<\/strong>\uff1aReMMDBench \u6709 500 samples\u30012,756 images\u30015-way L1 \u8207 8 \u985e L2 \u6a19\u7c64<\/li>\n<li><strong>\u76f8\u95dc\u6a21\u578b<\/strong>\uff1aQwen-family 4B \/ 9B \/ 27B\uff1b\u9996\u9801\u4ea6\u63d0\u5230 GPT-5.2 \u66fe\u7528\u65bc leaderboard<\/li>\n<li><strong>\u8f03\u9069\u5408\u7684\u60c5\u5883<\/strong>\uff1a\u7814\u7a76\u5718\u968a\u3001\u4e8b\u5be6\u67e5\u6838\u6d41\u7a0b\u8a2d\u8a08\u8005\u3001\u591a\u8a9e\u5167\u5bb9\u5be9\u6838\u8207 agent benchmark \u6bd4\u8f03<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u6027\u80fd\u65b9\u9762\uff0c\u5009\u5eab\u91cd\u9ede\u662f\u91cd\u73fe\u8ad6\u6587\u4e2d\u4e09\u5957\u7cfb\u7d71\u5728 500-sample ReMMDBench \u7684\u7d50\u679c\uff0c\u800c\u4e0d\u662f\u63d0\u4f9b\u4e00\u500b\u5373\u88dd\u5373\u7528\u7684\u7dda\u4e0a\u670d\u52d9\u3002\u5b83\u8f03\u9069\u5408\u62ff\u4f86\u505a benchmark \u9a57\u8b49\u3001\u5206\u6790\u4e0d\u540c agent pipeline \u7684\u8868\u73fe\uff0c\u6216\u8005\u7814\u7a76 evidence reuse \u5c0d\u591a\u6a21\u614b\u5224\u65b7\u6709\u5e7e\u5927\u5e6b\u52a9\uff1b\u8981\u76f4\u63a5\u653e\u9032\u7522\u54c1\uff0c\u4ecd\u8981\u81ea\u884c\u88dc\u56de\u8cc7\u6599\u63a5\u5165\u3001\u670d\u52d9\u5c01\u88dd\u8207\u66f4\u7a69\u5b9a\u7684\u63a8\u7406\u57fa\u5efa\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/DANG-ai\/ReMMDBench-Agent\" rel=\"noopener noreferrer\">https:\/\/github.com\/DANG-ai\/ReMMDBench-Agent<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9805\u76ee\u4e3b\u9801\uff1a<\/strong> <a href=\"https:\/\/dang-ai.github.io\/ReMMD\/\" rel=\"noopener noreferrer\">https:\/\/dang-ai.github.io\/ReMMD\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u662f\u4e00\u7d44\u91cd\u73fe\u8ad6\u6587\u7d50\u679c\u7684\u7814\u7a76\u4ee3\u78bc\uff0c\u96c6\u4e2d\u6bd4\u8f03\u4e09\u7a2e\u591a\u6a21\u614b misinformation detection agent \u7cfb\u7d71\u3002\u91cd\u9ede\u4e0d\u53ea\u770b\u5206\u6578\uff0c\u4ea6\u770b\u8b49\u64da\u5982\u4f55\u88ab\u62c6\u89e3\u3001\u6aa2\u7d22\u8207\u91cd\u7528\u3002<\/p>\n","protected":false},"author":8,"featured_media":9656,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[185,116,137,30,38,28,152,109,119,170,150,160,149,189,200,197,196],"tags":[],"class_list":["post-9657","post","type-post","status-publish","format-standard","hentry","category-qwen","category-agentic","category-api","category-image","category-38","category-services","category-python","category-rag","category-119","category-170","category-150","category-160","category-149","category-189","category-200","category-framework","category-196"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9657","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=9657"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9657\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9656"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9657"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9657"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9657"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}