{"id":22362,"date":"2020-12-05T22:31:08","date_gmt":"2020-12-05T14:31:08","guid":{"rendered":"https:\/\/swarma.org\/?p=22362"},"modified":"2020-12-05T22:31:08","modified_gmt":"2020-12-05T14:31:08","slug":"%e3%80%8a%e5%9b%a0%e6%9e%9c%e7%a7%91%e5%ad%a6%e5%91%a8%e5%88%8a%e3%80%8b%e7%ac%ac4%e6%9c%9f%ef%bc%9a%e5%9b%a0%e6%9e%9c%e8%b5%8b%e8%83%bd%e6%8e%a8%e8%8d%90%e7%b3%bb%e7%bb%9f","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=22362","title":{"rendered":"\u300a\u56e0\u679c\u79d1\u5b66\u5468\u520a\u300b\u7b2c4\u671f\uff1a\u56e0\u679c\u8d4b\u80fd\u63a8\u8350\u7cfb\u7edf"},"content":{"rendered":"
\n
\n
\n
\n

\u4e3a\u4e86\u5e2e\u52a9\u5927\u5bb6\u66f4\u597d\u5730\u4e86\u89e3\u56e0\u679c\u79d1\u5b66\u7684\u6700\u65b0\u79d1\u7814\u8fdb\u5c55\u548c\u8d44\u8baf\uff0c\u6211\u4eec\u56e0\u679c\u79d1\u5b66\u793e\u533a\u56e2\u961f<\/span>\u672c\u5468\u6574\u7406\u4e86\u7b2c4\u671f\u300a\u56e0\u679c\u79d1\u5b66\u5468\u520a\u300b\uff0c\u63a8\u9001\u8fd1\u671f\u56e0\u679c\u79d1\u5b66\u503c\u5f97\u5173\u6ce8\u7684\u8bba\u6587\u548c\u8d44\u8baf\u4fe1\u606f\uff0c \u540c\u65f6\u6211\u4eec\u4e5f\u5c06\u5411\u5927\u5bb6\u4ecb\u7ecd\u793e\u533a\u6b63\u5728\u63a8\u8fdb\u7684\u6d3b\u52a8\u2014\u2014\u56e0\u679c\u79d1\u5b66\u4e0eCausal AI\u8bfb\u4e66\u4f1a\u7b2c8\u671f\u4e2d\u7684\u4e3b\u8981\u62a5\u544a\u5185\u5bb9\u3001\u89c2\u70b9\u3002<\/p>\n

\u672c\u671f\u4f5c\u8005\uff1a\u4faf\u8339 \u95eb\u548c\u4e1c \u9648\u6657\u66e6 \u9f9a\u9e64\u626c \u9648\u5929\u8c6a \u6768\u4e8c\u8336<\/em><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n

<\/section>\n<\/section>\n
\u56e0\u679c\u79d1\u5b66\u793e\u533a\u7b80\u4ecb<\/strong>\uff1a\u5b83\u662f\u7531\u96c6\u667a\u4ff1\u4e50\u90e8<\/span>\u3001\u667a\u6e90\u793e\u533a\u5171\u540c\u63a8\u52a8\uff0c\u9762\u5411\u56e0\u679c\u79d1\u5b66\u9886\u57df\u7684\u5782\u76f4\u578b\u5b66\u672f\u8ba8\u8bba\u793e\u533a\uff0c\u76ee\u7684\u662f\u4fc3\u8fdb\u56e0\u679c\u79d1\u5b66\u4e13\u4e1a\u4eba\u58eb\u548c\u5174\u8da3\u7231\u597d\u8005\u4eec\u7684\u4ea4\u6d41\u548c\u5408\u4f5c\uff0c\u63a8\u8fdb\u56e0\u679c\u79d1\u5b66\u5b66\u672f\u3001\u4ea7\u4e1a\u751f\u6001\u7684\u5efa\u8bbe\u548c\u843d\u5730\uff0c\u5b55\u80b2\u65b0\u4e00\u4ee3\u56e0\u679c\u79d1\u5b66\u9886\u57df\u7684\u5b66\u672f\u4e13\u5bb6\u548c\u4ea7\u4e1a\u521b\u65b0\u8005\u3002<\/section>\n
   <\/span><\/section>\n
\u56e0\u679c\u79d1\u5b66\u793e\u533a\u6b22\u8fce\u60a8\u52a0\u5165\uff01<\/section>\n
 <\/span><\/section>\n

\u56e0\u679c\u79d1\u5b66\u793e\u533a\u613f\u666f<\/strong>\uff1a<\/em>\u56de\u7b54\u56e0\u679c\u95ee\u9898\u662f\u5404\u4e2a\u9886\u57df\u8feb\u5207\u7684\u9700\u6c42\uff0c\u5f53\u524d\u8bb8\u591a\u4e0d\u540c\u9886\u57df\uff08\u4f8b\u5982 AI \u548c\u7edf\u8ba1\u5b66\uff09<\/span>\u90fd\u5728\u4f7f\u7528\u56e0\u679c\u63a8\u7406\uff0c\u4f46\u662f\u4ed6\u4eec\u6240\u4f7f\u7528\u7684\u8bed\u8a00\u548c\u6a21\u578b\u5404\u4e0d\u76f8\u540c\uff0c\u5bfc\u81f4\u8fd9\u4e9b\u9886\u57df\u79d1\u5b66\u5bb6\u4e4b\u95f4\u6c9f\u901a\u4ea4\u6d41\u56f0\u96be\u3002\u56e0\u6b64\u6211\u4eec\u5e0c\u671b\u6784\u5efa\u4e00\u4e2a\u793e\u533a\uff0c\u901a\u8fc7\u7ec4\u7ec7\u5927\u91cf\u5b66\u672f\u6d3b\u52a8\uff0c\u4f7f\u5f97\u79d1\u7814\u4eba\u5458\u80fd\u591f\u638c\u63e1\u7edf\u8ba1\u5b66\u7684\u6838\u5fc3\u601d\u60f3\uff0c\u719f\u7ec3\u4f7f\u7528\u5f53\u524d AI \u5404\u79cd\u6280\u672f\uff08\u4f8b\u5982 Pytorch\/Pyro \u642d\u5efa\u6df1\u5ea6\u6982\u7387\u6a21\u578b\uff09<\/span>\uff0c\u4fc3\u8fdb\u5404\u4e2a\u9886\u57df\u7684\u7814\u7a76\u8005\u4ea4\u6d41\u548c\u601d\u7ef4\u78b0\u649e\uff0c\u4ece\u800c\u8ba9\u5404\u4e2a\u9886\u57df\u7684\u56e0\u679c\u63a8\u7406\u6709\u7740\u5171\u540c\u7684\u8303\u5f0f\uff0c\u751a\u81f3\u662f\u5171\u540c\u7684\u5de5\u7a0b\u5b9e\u8df5\u6807\u51c6\uff0c\u63a8\u52a8\u521a\u521a\u6210\u578b\u7684\u56e0\u679c\u79d1\u5b66\u5feb\u901f\u5411\u524d\u53d1\u5c55\u3002<\/strong>\u5177\u5907\u56e0\u679c\u63a8\u7406\u80fd\u529b\u7684\u4eba\u7c7b\u7d27\u5bc6\u534f\u4f5c\u521b\u9020\u4e86\u5f3a\u5927\u7684\u6587\u660e\uff0c\u6211\u4eec\u5e0c\u671b\u5728\u672a\u6765\u793e\u4f1a\u4e2d\uff0c\u56e0\u679c\u63a8\u7406\u878d\u5165\u5230\u6bcf\u4e2a\u5b66\u79d1\uff0c\u5c24\u5176\u662f\u7d27\u5bc6\u7ed3\u5408\u548c\u63d0\u5347 AI \uff0c\u671f\u5f85\u65e0\u6570\u5177\u5907\u6500\u767b\u56e0\u679c\u4e4b\u68af\u80fd\u529b\u7684 Agents (Causal AI) \u548c\u4eba\u7c7b\u4e00\u8d77\u534f\u4f5c\uff0c\u5171\u5efa\u4e0b\u4e00\u4ee3\u7684\u4eba\u7c7b\u6587\u660e\uff01<\/strong><\/p>\n

 <\/p>\n

\u5982\u679c\u60a8\u6709\u9002\u5f53\u7684\u6570\u5b66\u57fa\u7840\u548c\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u7ecf\u9a8c\uff0c\u65e2\u6709\u79d1\u5b66\u5bb6\u7684\u597d\u5947\u5fc3\u4e5f\u6709\u5de5\u7a0b\u5e08\u601d\u7ef4\uff0c\u5e0c\u671b\u53c2\u4e0e\u5230\u201d\u56e0\u679c\u9769\u547d\u201c\u4e2d\uff0c\u6559\u4f1a\u673a\u5668\u56e0\u679c\u601d\u7ef4\uff0c\u4e3a\u56e0\u679c\u79d1\u5b66\u4f5c\u51fa\u8d21\u732e\uff0c\u8bf7\u52a0\u5165\u6211\u4eec\u5fae\u4fe1\u7fa4\uff1a\u626b\u63cf\u4e0b\u9762\u793e\u533a\u5c0f\u52a9\u624b\u4e8c\u7ef4\u7801\u52a0\u5165\uff08\u8bf7\u5907\u6ce8\u201c\u56e0\u679c\u79d1\u5b66\u201d\uff09\ud83d\udc47<\/span><\/strong><\/p>\n


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

        <\/section>\n

<\/section>\n

\u5f53\u4eca\u7684\u4fe1\u606f\u9769\u547d\u65f6\u4ee3\uff0c\u4fe1\u606f\u5904\u7406\u548c\u5229\u7528\u80fd\u529b\u662f AI \u667a\u80fd\u6c34\u5e73\u7684\u4e00\u4e2a\u91cd\u8981\u65b9\u9762\u3002\u800c\u56e0\u679c\u63a8\u7406\u80fd\u591f\u5e2e\u52a9\u6211\u4eec\u5229\u7528 Lorenzian imagined space \u4e2d\u4fe1\u606f\u8fdb\u884c\u51b3\u7b56<\/span> (Sch\u00f6lkopf B., 2019)<\/span>\uff0c\u56e0\u679c\u63a8\u7406\u662f\u4e0b\u4e00\u4ee3 AI \u7684\u5fc5\u8981\u7ec4\u4ef6<\/span> (Pearl J., 2019, Elias B., 2020)<\/span>\u3002\u63a8\u8350\u662f\u4fe1\u606f\u63d0\u53d6\u7684\u4e00\u79cd\u7279\u6b8a\u5f62\u5f0f\uff0c\u5b83\u5229\u7528\u8fc7\u53bb\u7684\u884c\u4e3a\u548c\u7528\u6237\u76f8\u4f3c\u6027\u6765\u751f\u6210\u4e00\u7cfb\u5217\u4fe1\u606f\uff0c\u8fd9\u4e9b\u4fe1\u606f\u662f\u6839\u636e\u6700\u7ec8\u7528\u6237\u7684\u559c\u597d\u91cf\u8eab\u5b9a\u5236\u7684\u3002<\/span><\/p>\n


<\/section>\n
<\/section>\n
\u56fe1\uff1a\u63a8\u8350\u7cfb\u7edf\u7684\u89d2\u8272(Ricardo Baeza-Yates)<\/span><\/section>\n

<\/section>\n
\u5728\u6700\u8fd1\u4e3e\u884c\u7684\u63a8\u8350\u7cfb\u7edf\u6700\u91cd\u8981\u7684\u4f1a\u8bae RecSys 2020 \u4e2d\uff0c\u53ef\u4ee5\u770b\u5230\u5b66\u672f\u754c\u548c\u5de5\u4e1a\u754c\u4e00\u4e2a\u91cd\u8981\u8d8b\u52bf\u662f\u6709\u5173\u7684 bias \u7684\u7814\u7a76\uff0c\u4e5f\u5c31\u662f\u63a8\u8350\u7cfb\u7edf\u51fa\u73b0\u7684\u5404\u79cd\u504f\u5dee\u8ba9\u5176\u63a8\u8350\u975e\u9884\u671f\u7684 Item\u3002\u63a8\u8350\u7cfb\u7edf\u7684\u4efb\u52a1\u88ab\u7ecf\u5178\u5730\u5b9a\u4e49\u4e3a\u9884\u6d4b\u7528\u6237\u7684\u504f\u597d\u548c\u7528\u6237\u8bc4\u5206\u3002\u7136\u800c\uff0c\u5b83\u672c\u8d28\u4e0a\u662f\u8981\u56de\u7b54\u4e00\u4e2a\u53cd\u4e8b\u5b9e\u95ee\u9898\uff1a\u201c\u5982\u679c\u6211\u4eec\u2018\u5f3a\u8feb\u2019\u7528\u6237\u53bb\u770b\u7535\u5f71\uff0c\u8bc4\u5206\u4f1a\u662f\u591a\u5c11?\u201d \u5982\u4f55\u4f7f\u7528\u89c2\u6d4b\u6570\u636e\u6b63\u786e\u7684\u8bad\u7ec3\u63a8\u8350\u7b97\u6cd5\uff0c\u751a\u81f3\u8bc4\u4f30 (NOT A\/B \u6d4b\u8bd5) <\/span>\u8bc4\u4f30\u63a8\u8350\u7cfb\u7edf\u7684\u6027\u80fd\uff0c\u4f7f\u7528\u56e0\u679c\u63a8\u7406 debias \u6210\u4e3a\u5176\u4e2d\u4e00\u4e2a\u975e\u5e38 Promising \u7684\u65b9\u5411 (Wang Y., 2019, Sharma A., 2015)<\/span>\u3002<\/section>\n

<\/section>\n
<\/section>\n
\u56fe2: \u56e0\u679c\u63a8\u8350\u7684\u57fa\u672c\u6570\u5b66\u5f62\u5f0f<\/span><\/section>\n

<\/section>\n

\u6b64\u6b21\u6211\u4eec\u9080\u8bf7\u4e9a\u5229\u6851\u90a3\u5dde\u7acb\u5927\u5b66\u535a\u58eb\u90ed\u82e5\u57ce<\/span>\u63a8\u8350\u4e86\u4e00\u4e9b\u6709\u5173 Causal + RecSyc \u7684\u524d\u6cbf\u8bba\u6587\uff0c\u4e0b\u9762\u662f\u6211\u4eec\u7684\u6574\u7406\uff1a<\/p>\n

 <\/span><\/section>\n

<\/span><\/section>\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

1. \u8bba\u6587\u63a8\u8350<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/section>\n
1.1 Unbiased learning to rank<\/span><\/strong><\/span><\/section>\n

<\/section>\n
Joachims, Thorsten, Adith Swaminathan, and Tobias     Schnabel. “Unbiased learning-to-rank with biased feedback.” In Proceedings of the Tenth ACM     International Conference on Web Search and Data Mining, pp. 781-789.     2017.<\/span><\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n
\u5728\u4eba\u673a\u4ea4\u4e92\u7cfb\u7edf\u4e2d\uff0c\u9690\u5f0f\u53cd\u9988<\/span>\uff08\u4f8b\u5982\u70b9\u51fb\u91cf\u3001\u505c\u7559\u65f6\u95f4\u7b49\uff09<\/span>\u662f\u4e00\u4e2a\u4e30\u5bcc\u7684\u6570\u636e\u6765\u6e90\u3002\u867d\u7136\u9690\u5f0f\u53cd\u9988\u6709\u5f88\u591a\u4f18\u52bf<\/span>\uff08\u4f8b\u5982\uff0c\u641c\u96c6\u6210\u672c\u4f4e\u3001\u4ee5\u7528\u6237\u4e3a\u4e2d\u5fc3\u3001\u53ca\u65f6\uff09<\/span>\uff0c\u5b83\u7684\u56fa\u6709\u504f\u5dee\u4e0d\u5229\u4e8e\u5b83\u7684\u6709\u6548\u4f7f\u7528\u3002\u4f8b\u5982\uff0c\u5728\u641c\u7d22\u6392\u5e8f\u4e2d\uff0c\u67d0\u4e2a\u7ed3\u679c\u7684\u4f4d\u7f6e\u504f\u5dee\u5f3a\u70c8\u5f71\u54cd\u5b83\u7684\u70b9\u51fb\u91cf\uff0c\u56e0\u6b64\uff0c\u76f4\u63a5\u4f7f\u7528\u70b9\u51fb\u91cf\u4f5c\u4e3aLearning-to-Rank\u65b9\u6cd5<\/span>\uff08\u8bd1\u8005\u6ce8\uff1a\u4ee5\u4e0b\u7b80\u79f0LTR\uff09<\/span>\u7684\u8bad\u7ec3\u6570\u636e\u4f1a\u4ea7\u751f\u6b21\u4f18\u7ed3\u679c\u3002\u4e3a\u4e86\u514b\u670d\u8fd9\u79cd\u504f\u5dee\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u53cd\u4e8b\u5b9e\u63a8\u7406\u6846\u67b6\uff0c\u5373\u4f7f\u5728\u6709\u504f\u6570\u636e\u4e0b\uff0c\u5b83\u4e5f\u80fd\u901a\u8fc7\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316\u4e3a\u65e0\u504f\u7684LTR\u63d0\u4f9b\u7406\u8bba\u57fa\u7840\u3002\u4f7f\u7528\u8fd9\u4e2a\u6846\u67b6\uff0c\u6211\u4eec\u63a8\u5bfc\u4e86\u503e\u5411\u52a0\u6743\u6392\u5e8f\u7684\u652f\u6301\u5411\u91cf\u673a\u7528\u4e8e\u9690\u5f0f\u53cd\u9988\u7684\u5224\u522b\u5f0f\u5b66\u4e60\uff0c\u8fd9\u91cc\u70b9\u51fb\u91cf\u6a21\u578b\u88ab\u7528\u6765\u4f30\u8ba1\u503e\u5411\u6027\u3002\u4e0d\u540c\u4e8e\u5927\u90e8\u5206\u4f20\u7edf\u7684\u4f7f\u7528\u70b9\u51fb\u91cf\u6a21\u578b\u53bb\u9664\u6570\u636e\u504f\u5dee\u7684\u65b9\u6cd5\uff0c\u5728\u6ca1\u6709\u91cd\u590d\u67e5\u8be2\u7684\u60c5\u51b5\u4e0b\uff0c\u6211\u4eec\u7684\u65b9\u6cd5\u4f9d\u7136\u80fd\u591f\u8bad\u7ec3\u6392\u5e8f\u51fd\u6570\u3002\u9664\u4e86\u7406\u8bba\u63a8\u5bfc\uff0c\u5b9e\u8bc1\u5206\u6790\u8868\u660e\u6211\u4eec\u63d0\u51fa\u7684\u5b66\u4e60\u65b9\u6cd5\u5bf9\u5904\u7406\u504f\u5dee\u975e\u5e38\u6709\u6548\u3001\u5bf9\u566a\u58f0\u548c\u503e\u5411\u6a21\u578b\u7684\u8bef\u8bc6\u522b\u6709\u7a33\u5065\u6027\u3001\u5e76\u4e14\u662f\u6709\u6548\u7387\u7684\u3002\u6211\u4eec\u4e5f\u5c55\u793a\u4e86\u8fd9\u4e2a\u65b9\u6cd5\u5728\u73b0\u5b9e\u4e16\u754c\u7684\u9002\u7528\u6027\uff0c\u5728\u8fd0\u884c\u7684\u641c\u7d22\u5f15\u64ce\u4e0a\uff0c\u5b83\u80fd\u591f\u6301\u7eed\u6539\u5584\u68c0\u7d22\u6c34\u5e73\u3002<\/span><\/section>\n

<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u4faf\u8339<\/span><\/section>\n

<\/section>\n
Wang, Xuanhui, Nadav Golbandi, Michael Bendersky,     Donald Metzler, and Marc Najork. “Position bias estimation for     unbiased learning to rank in personal search.” In Proceedings of the Eleventh ACM     International Conference on Web Search and Data Mining, pp. 610-618.     2018.<\/span>
<\/section>\n

<\/section>\n

\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong>\u4f17\u6240\u5468\u77e5\uff0c\u70b9\u51fb\u91cf\u5b66\u4e60\u7684\u6311\u6218\u662f\u5b83\u7684\u56fa\u6709\u504f\u5dee\uff0c\u5c24\u5176\u662f\u4f4d\u7f6e\u504f\u5dee\u3002\u4f20\u7edf\u7684\u70b9\u51fb\u91cf\u6a21\u578b\u65e8\u5728\u4ece\u67e5\u8be2\u548c\u6587\u6863\u76f8\u5173\u6027\u63d0\u53d6\u4fe1\u606f\uff0c\u5728\u63d0\u53d6\u76f8\u5173\u6027\u4fe1\u606f\u4e4b\u540e\u5c31\u4f1a\u53bb\u6389\u4f30\u8ba1\u7684\u504f\u5dee\u3002\u4e0d\u540c\u4e8e\u6b64\uff0c\u6700\u65b0\u7684\u65e0\u504fLearning-to-Rank\u65b9\u6cd5\u80fd\u591f\u6709\u6548\u5730\u5229\u7528\u504f\u5dee\uff0c\u8fdb\u800c\u7740\u773c\u4e8e\u4f30\u8ba1\u504f\u5dee\u800c\u975e\u76f8\u5173\u6027[20, 31]\u3002\u73b0\u6709\u7684\u65b9\u6cd5\u5728\u5c0f\u8303\u56f4\u7684\u4ea7\u54c1\u6d41\u91cf\u4e2d\u968f\u673a\u5316\u641c\u7d22\u7ed3\u679c\u6765\u4f30\u8ba1\u4f4d\u7f6e\u504f\u5dee\u3002\u8fd9\u4e0d\u80fd\u8fbe\u5230\u7406\u60f3\u7ed3\u679c\uff0c\u56e0\u4e3a\u7ed3\u679c\u7684\u968f\u673a\u5316\u5bf9\u7528\u6237\u7684\u641c\u7d22\u4f53\u9a8c\u6709\u4e0d\u597d\u7684\u5f71\u54cd\u3002\u672c\u6587\u6bd4\u8f83\u4e86\u4e0d\u540c\u7684\u7ed3\u679c\u968f\u673a\u5316\u65b9\u6cd5\u5e76\u5c55\u793a\u4e86\u5b83\u4eec\u5bf9\u4e8e\u7528\u6237\u641c\u7d22\u7684\u4e0d\u5229\u5f71\u54cd\u3002\u7136\u540e\uff0c\u6211\u4eec\u7814\u7a76\u4e86\u5982\u4f55\u5728\u4e0d\u4f9d\u8d56\u968f\u673a\u5316\u7684\u60c5\u51b5\u4e0b\uff0c\u4ece\u5e38\u89c4\u7684\u70b9\u51fb\u91cf\u6570\u636e\u63a8\u65ad\u4f4d\u7f6e\u504f\u5dee\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u57fa\u4e8e\u56de\u5f52\u7684\u671f\u671b\u6700\u5927\u5316\u7b97\u6cd5\uff0c\u5728\u4f4d\u7f6e\u504f\u5dee\u70b9\u51fb\u91cf\u6a21\u578b\u7684\u57fa\u7840\u4e0a\uff0c\u80fd\u591f\u5904\u7406\u7528\u6237\u641c\u7d22\u4e2d\u7684\u5927\u91cf\u7a00\u758f\u6570\u636e\u3002\u6211\u4eec\u8bc4\u4f30\u4e86\u6211\u4eec\u7684\u671f\u671b\u6700\u5927\u5316\u7b97\u6cd5\u548cLearning-to-Rank\u65b9\u6cd5\u63d0\u53d6\u7684\u504f\u5dee\u3002\u6211\u4eec\u7684\u7ed3\u679c\u8868\u660e\uff0c\u4e0d\u4f7f\u7528\u7ed3\u679c\u968f\u673a\u5316\u800c\u76f4\u63a5\u4ece\u5e38\u89c4\u7684\u70b9\u51fb\u91cf\u63d0\u53d6\u4f4d\u7f6e\u504f\u5dee\u662f\u6709\u524d\u666f\u7684\u3002\u63d0\u53d6\u7684\u504f\u5dee\u80fd\u663e\u8457\u6539\u5584Learning-to-Rank\u7b97\u6cd5\u3002\u8fdb\u4e00\u6b65\uff0c\u6211\u4eec\u6bd4\u8f83\u4e86\u9010\u70b9\u548c\u9010\u5bf9\u7684Learning-to-Rank\u6a21\u578b\u3002\u7ed3\u679c\u8868\u660e\u9010\u5bf9\u7684\u6a21\u578b\u80fd\u66f4\u6709\u6548\u5730\u5229\u7528\u4f30\u8ba1\u7684\u504f\u5dee\u3002<\/span>
<\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u4faf\u8339<\/span><\/section>\n

<\/section>\n
Ai, Qingyao, Keping Bi, Cheng Luo, Jiafeng Guo, and W. Bruce Croft. “Unbiased learning to rank with unbiased propensity estimation.” In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, pp. 385-394. 2018.<\/span>
<\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong>
<\/section>\n

\u7528\u6709\u504f\u7684\u70b9\u51fb\u6570\u636e\u5b66\u4e60\u5982\u4f55\u6392\u540d\u662f\u4f17\u6240\u5468\u77e5\u7684\u6311\u6218\u3002\u4e3a\u4e86\u5b66\u4e60\u5982\u4f55\u6392\u540d\uff0c\u4eba\u4eec\u63a2\u7d22\u51fa\u591a\u79cd\u65b9\u6cd5\u53bb\u9664\u70b9\u51fb\u6570\u636e\u4e2d\u7684\u504f\u5dee\uff0c\u5982\u70b9\u51fb\u6a21\u578b\u3001\u7ed3\u679c\u4ea4\u9519\uff0c\u8fd8\u6709\u6700\u8fd1\u7684\u57fa\u4e8e\u53cd\u503e\u5411\u52a0\u6743<\/span>\uff08IPW\uff09<\/span>\u7684\u5b66\u4e60\u5982\u4f55\u6392\u540d\u7684\u65e0\u504f\u6846\u67b6\u3002\u5ffd\u7565\u5b83\u4eec\u4e4b\u95f4\u7684\u4e0d\u540c\u70b9\uff0c\u6700\u8fd1\u5927\u591a\u6570\u7814\u7a76\u81f4\u529b\u4e8e\u4ece\u6392\u540d\u7b97\u6cd5\u7684\u5b66\u4e60\u4e2d\u5355\u72ec\u4f30\u8ba1\u70b9\u51fb\u504f\u5dee<\/span>\uff08\u79f0\u4e3a\u503e\u5411\u6a21\u578b\uff09<\/span>\u3002\u4e3a\u4e86\u4f30\u8ba1\u70b9\u51fb\u503e\u5411\uff0c\u4ed6\u4eec\u6709\u7684\u8fdb\u884c\u4e86\u5728\u7ebf\u7ed3\u679c\u968f\u673a\u5316\uff0c\u4f46\u8fd9\u4f1a\u5f71\u54cd\u7528\u6237\u4f53\u9a8c\uff1b\u6709\u7684\u8fdb\u884c\u79bb\u7ebf\u53c2\u6570\u4f30\u8ba1\uff0c\u4f46\u5bf9\u70b9\u51fb\u6570\u636e\u6709\u7279\u6b8a\u8981\u6c42\uff0c\u800c\u4e14\u4f18\u5316\u76ee\u6807<\/span>\uff08\u5982\u70b9\u51fb\u4f3c\u7136\uff09<\/span>\u4e0d\u662f\u76f4\u63a5\u548c\u6392\u540d\u7cfb\u7edf\u7684\u6027\u80fd\u6709\u5173\u3002\u5728\u8fd9\u4e2a\u5de5\u4f5c\u4e2d\uff0c\u6211\u4eec\u901a\u8fc7\u7ed3\u5408\u503e\u5411\u6a21\u578b\u548c\u6392\u540d\u6a21\u578b\u6765\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\u3002\u6211\u4eec\u53d1\u73b0\u4ece\u70b9\u51fb\u6570\u636e\u4e2d\u4f30\u8ba1\u4e00\u4e2a\u503e\u5411\u6a21\u578b\u662f\u5b66\u4e60\u65e0\u504f\u6392\u540d\u7684\u4e00\u4e2a\u5bf9\u5076\u95ee\u9898\u3002\u57fa\u4e8e\u8fd9\u70b9\u89c2\u5bdf\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u5bf9\u5076\u5b66\u4e60\u7b97\u6cd5<\/span>\uff08DLA\uff09<\/span>\uff0c\u5b83\u53ef\u4ee5\u5b66\u4e60\u4e00\u4e2a\u65e0\u504f\u7684\u6392\u540d\u5668\u548c\u4e00\u4e2a\u65e0\u504f\u7684\u503e\u5411\u6a21\u578b\u3002DLA\u662f\u81ea\u52a8\u5730\u65e0\u504f\u5730\u5b66\u4e60\u6392\u540d\u7684\u6846\u67b6\uff0c\u56e0\u4e3a\u5b83\u4ece\u672a\u8fdb\u884c\u9884\u5904\u7406\u7684\u6709\u504f\u70b9\u51fb\u6570\u636e\u4e2d\u76f4\u63a5\u5b66\u4e60\u65e0\u504f\u6392\u540d\u6a21\u578b\u3002\u5b83\u53ef\u4ee5\u9002\u5e94\u504f\u5dee\u5206\u5e03\u7684\u53d8\u5316\uff0c\u53ef\u4ee5\u5e94\u7528\u5230\u5728\u7ebf\u5b66\u4e60\u4e2d\u3002\u6211\u4eec\u7684\u5b9e\u9a8c\u4f7f\u7528\u4e86\u5408\u6210\u7684\u6570\u636e\u96c6\u548c\u771f\u5b9e\u4e16\u754c\u7684\u6570\u636e\u96c6\uff0c\u7ed3\u679c\u663e\u793a\uff1a\u7528DLA\u8bad\u7ec3\u6a21\u578b\u7684\u6027\u80fd\u660e\u663e\u8d85\u8fc7\u4e86\u57fa\u4e8e\u7ed3\u679c\u968f\u673a\u5316\u7684\u5b66\u4e60\u6392\u540d\u7684\u7b97\u6cd5\uff0c\u548c\u901a\u8fc7\u7528\u4ece\u70b9\u51fb\u6a21\u578b\u4e2d\u63d0\u53d6\u7684\u76f8\u5173\u6027\u4fe1\u53f7\u8bad\u7ec3\u7684\u6a21\u578b\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u95eb\u548c\u4e1c<\/span><\/section>\n

<\/span><\/section>\n
Guo, Ruocheng, Xiaoting Zhao, Adam Henderson, Liangjie Hong, and Huan Liu. “Debiasing Grid-based Product Search in E-commerce.” In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 2852-2860. 2020.<\/span>
<\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u968f\u7740\u7535\u5b50\u5546\u52a1\u5728\u65e5\u5e38\u751f\u6d3b\u7684\u5e7f\u6cdb\u4f7f\u7528\uff0c\u4e30\u5bcc\u7684\u9690\u5f0f\u53cd\u9988\u6570\u636e\u4e3a\u8bad\u7ec3\u548c\u6d4b\u8bd5\u7f51\u8d2d\u641c\u7d22\u6392\u5e8f\u7b97\u6cd5\u63d0\u4f9b\u4e86\u57fa\u7840\u3002\u5c3d\u7ba1\u4fbf\u4e8e\u641c\u96c6\uff0c\u9690\u5f0f\u53cd\u9988\u6570\u636e\u6709\u591a\u79cd\u56fa\u6709\u504f\u5dee\uff0c\u56e0\u4e3a\u73b0\u6709\u7684\u641c\u7d22\u6392\u5e8f\u7b97\u6cd5\u4f7f\u5f97\u7528\u6237\u7684\u53cd\u9988\u4ec5\u5c40\u9650\u5728\u63a5\u89e6\u5230\u7684\u4ea7\u54c1\uff0c\u5e76\u4e14\u53d7\u4ea7\u54c1\u5c55\u793a\u65b9\u5f0f\u7684\u5f71\u54cd\u3002\u5927\u90e8\u5206\u73b0\u6709\u7684\u65b9\u6cd5\u5728\u57fa\u4e8e\u5217\u8868\u7684\u7f51\u9875\u641c\u7d22\u573a\u666f\u4e2d\u5b9e\u73b0\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u4e0d\u80fd\u76f4\u63a5\u7528\u4e8e\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\uff0c\u6709\u4e0b\u8ff0\u4e24\u4e2a\u539f\u56e0\u3002\u7b2c\u4e00\uff0c\u5728\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\u4e2d\uff0c\u641c\u7d22\u5f15\u64ce\u7ed3\u679c\u9875\u9762\u4ee52\u7ef4\u683c\u70b9\u8868\u793a\uff0c\u800c\u73b0\u6709\u7684\u65b9\u6cd5\u5e76\u4e0d\u8003\u8651\u57fa\u4e8e\u5217\u8868\u7684\u7f51\u9875\u641c\u7d22\u548c\u57fa\u4e8e\u683c\u70b9\u7684\u4ea7\u54c1\u641c\u7d22\u7684\u7528\u6237\u884c\u4e3a\u5dee\u5f02\u3002\u7b2c\u4e8c\uff0c\u5728\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\u6709\u591a\u79cd\u7c7b\u578b\u7684\u9690\u5f0f\u53cd\u9988<\/span>\uff08\u4f8b\u5982\uff0c\u70b9\u51fb\u548c\u8d2d\u4e70\uff09<\/span>\uff0c\u6211\u4eec\u7684\u76ee\u6807\u662f\u5c06\u6240\u6709\u7c7b\u578b\u7684\u9690\u5f0f\u53cd\u9988\u4f5c\u4e3a\u76d1\u7763\u4fe1\u53f7\u3002\u672c\u6587\u8003\u8651\u4e86\u57fa\u4e8e\u683c\u70b9\u7684\u4ea7\u54c1\u641c\u7d22\u573a\u666f\uff0c\u5c06\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u62d3\u5c55\u5230\u7535\u5b50\u5546\u52a1\u641c\u7d22\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u65b0\u9896\u7684\u6846\u67b6\uff0c\u4e00\u65b9\u9762\u5f62\u6210\u4e86\u5728\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u4e2d\u4f7f\u7528\u591a\u79cd\u9690\u5f0f\u53cd\u9988\u7684\u7406\u8bba\u57fa\u7840\uff0c\u53e6\u4e00\u65b9\u9762\u5305\u542b\u4e86\u884c\u7701\u7565\u548c\u7f13\u6162\u8870\u51cf\u7684\u70b9\u51fb\u6a21\u578b\uff0c\u5728\u57fa\u4e8e\u683c\u70b9\u4ea7\u54c1\u641c\u7d22\u7684\u9006\u5411\u503e\u5411\u8bc4\u5206\u4e2d\u6293\u53d6\u72ec\u7279\u7684\u7528\u6237\u884c\u4e3a\u6a21\u5f0f\u3002\u5728\u5927\u91cf\u73b0\u5b9e\u4e16\u754c\u4e2d\u4e0d\u540c\u6d4f\u89c8\u8bbe\u5907\u548c\u4ea7\u54c1\u7684\u7535\u5b50\u5546\u52a1\u641c\u7d22\u8bb0\u5f55\u6570\u636e\u96c6\u4e0a\u7684\u8bd5\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u63d0\u51fa\u7684\u6846\u67b6\u4f18\u4e8e\u6700\u5148\u8fdb\u7684\u65e0\u504f\u6392\u5e8f\u5b66\u4e60\u7b97\u6cd5\u3002\u8fd9\u4e9b\u7ed3\u679c\u540c\u65f6\u63ed\u793a\u4e86\u5728\u4e0d\u540c\u6d4f\u89c8\u8bbe\u5907\u548c\u4ea7\u54c1\u4e2d\u7528\u6237\u884c\u4e3a\u6a21\u5f0f\u5982\u4f55\u53d8\u5316\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u4faf\u8339<\/span><\/section>\n

<\/section>\n
1.3 Debiasing Recommendation systems<\/span><\/strong><\/section>\n

<\/span><\/section>\n
Guo, Ruocheng, Xiaoting Zhao, Adam Henderson, Liangjie Hong, and Huan Liu. “Debiasing Grid-based Product Search in E-commerce.” In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 2852-2860. 2020.<\/span>
<\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u968f\u7740\u7535\u5b50\u5546\u52a1\u5728\u65e5\u5e38\u751f\u6d3b\u7684\u5e7f\u6cdb\u4f7f\u7528\uff0c\u4e30\u5bcc\u7684\u9690\u5f0f\u53cd\u9988\u6570\u636e\u4e3a\u8bad\u7ec3\u548c\u6d4b\u8bd5\u7f51\u8d2d\u641c\u7d22\u6392\u5e8f\u7b97\u6cd5\u63d0\u4f9b\u4e86\u57fa\u7840\u3002\u5c3d\u7ba1\u4fbf\u4e8e\u641c\u96c6\uff0c\u9690\u5f0f\u53cd\u9988\u6570\u636e\u6709\u591a\u79cd\u56fa\u6709\u504f\u5dee\uff0c\u56e0\u4e3a\u73b0\u6709\u7684\u641c\u7d22\u6392\u5e8f\u7b97\u6cd5\u4f7f\u5f97\u7528\u6237\u7684\u53cd\u9988\u4ec5\u5c40\u9650\u5728\u63a5\u89e6\u5230\u7684\u4ea7\u54c1\uff0c\u5e76\u4e14\u53d7\u4ea7\u54c1\u5c55\u793a\u65b9\u5f0f\u7684\u5f71\u54cd\u3002\u5927\u90e8\u5206\u73b0\u6709\u7684\u65b9\u6cd5\u5728\u57fa\u4e8e\u5217\u8868\u7684\u7f51\u9875\u641c\u7d22\u573a\u666f\u4e2d\u5b9e\u73b0\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u4e0d\u80fd\u76f4\u63a5\u7528\u4e8e\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\uff0c\u6709\u4e0b\u8ff0\u4e24\u4e2a\u539f\u56e0\u3002\u7b2c\u4e00\uff0c\u5728\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\u4e2d\uff0c\u641c\u7d22\u5f15\u64ce\u7ed3\u679c\u9875\u9762\u4ee52\u7ef4\u683c\u70b9\u8868\u793a\uff0c\u800c\u73b0\u6709\u7684\u65b9\u6cd5\u5e76\u4e0d\u8003\u8651\u57fa\u4e8e\u5217\u8868\u7684\u7f51\u9875\u641c\u7d22\u548c\u57fa\u4e8e\u683c\u70b9\u7684\u4ea7\u54c1\u641c\u7d22\u7684\u7528\u6237\u884c\u4e3a\u5dee\u5f02\u3002\u7b2c\u4e8c\uff0c\u5728\u7535\u5b50\u5546\u52a1\u7f51\u7ad9\u6709\u591a\u79cd\u7c7b\u578b\u7684\u9690\u5f0f\u53cd\u9988<\/span>\uff08\u4f8b\u5982\uff0c\u70b9\u51fb\u548c\u8d2d\u4e70\uff09<\/span>\uff0c\u6211\u4eec\u7684\u76ee\u6807\u662f\u5c06\u6240\u6709\u7c7b\u578b\u7684\u9690\u5f0f\u53cd\u9988\u4f5c\u4e3a\u76d1\u7763\u4fe1\u53f7\u3002\u672c\u6587\u8003\u8651\u4e86\u57fa\u4e8e\u683c\u70b9\u7684\u4ea7\u54c1\u641c\u7d22\u573a\u666f\uff0c\u5c06\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u62d3\u5c55\u5230\u7535\u5b50\u5546\u52a1\u641c\u7d22\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u65b0\u9896\u7684\u6846\u67b6\uff0c\u4e00\u65b9\u9762\u5f62\u6210\u4e86\u5728\u65e0\u504f\u7684\u6392\u5e8f\u5b66\u4e60\u4e2d\u4f7f\u7528\u591a\u79cd\u9690\u5f0f\u53cd\u9988\u7684\u7406\u8bba\u57fa\u7840\uff0c\u53e6\u4e00\u65b9\u9762\u5305\u542b\u4e86\u884c\u7701\u7565\u548c\u7f13\u6162\u8870\u51cf\u7684\u70b9\u51fb\u6a21\u578b\uff0c\u5728\u57fa\u4e8e\u683c\u70b9\u4ea7\u54c1\u641c\u7d22\u7684\u9006\u5411\u503e\u5411\u8bc4\u5206\u4e2d\u6293\u53d6\u72ec\u7279\u7684\u7528\u6237\u884c\u4e3a\u6a21\u5f0f\u3002\u5728\u5927\u91cf\u73b0\u5b9e\u4e16\u754c\u4e2d\u4e0d\u540c\u6d4f\u89c8\u8bbe\u5907\u548c\u4ea7\u54c1\u7684\u7535\u5b50\u5546\u52a1\u641c\u7d22\u8bb0\u5f55\u6570\u636e\u96c6\u4e0a\u7684\u8bd5\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u63d0\u51fa\u7684\u6846\u67b6\u4f18\u4e8e\u6700\u5148\u8fdb\u7684\u65e0\u504f\u6392\u5e8f\u5b66\u4e60\u7b97\u6cd5\u3002\u8fd9\u4e9b\u7ed3\u679c\u540c\u65f6\u63ed\u793a\u4e86\u5728\u4e0d\u540c\u6d4f\u89c8\u8bbe\u5907\u548c\u4ea7\u54c1\u4e2d\u7528\u6237\u884c\u4e3a\u6a21\u5f0f\u5982\u4f55\u53d8\u5316\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u4faf\u8339<\/span><\/section>\n

<\/section>\n
Schnabel, Tobias, Adith Swaminathan, Ashudeep Singh,     Navin Chandak, and Thorsten Joachims. “Recommendations as treatments:     Debiasing learning and evaluation.” ICML 2016..<\/span><\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u5927\u90e8\u5206\u63a8\u8350\u7cfb\u7edf\u7684\u8bc4\u4f30\u548c\u8bad\u7ec3\u6570\u636e\u662f\u53d7\u9650\u4e8e\u9009\u62e9\u504f\u5dee\u7684\uff0c\u8fd9\u79cd\u504f\u5dee\u8981\u4e48\u901a\u8fc7\u7528\u6237\u81ea\u8eab\u9009\u62e9\u4ea7\u751f\u3001\u8981\u4e48\u901a\u8fc7\u63a8\u8350\u7cfb\u7edf\u7684\u884c\u4e3a\u4ea7\u751f\u3002\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e00\u79cd\u901a\u8fc7\u56e0\u679c\u63a8\u65ad\u8c03\u6574\u6a21\u578b\u548c\u4f30\u8ba1\u5668\u7684\u539f\u5219\u6027\u65b9\u6cd5\u6765\u89e3\u51b3\u9009\u62e9\u504f\u5dee\u7684\u95ee\u9898\u3002\u8fd9\u79cd\u65b9\u6cd5\u53ef\u4ee5\u5728\u6709\u504f\u6570\u636e\u4e0a\u5bfc\u51fa\u65e0\u504f\u7684\u6027\u80fd\u4f30\u8ba1\u5668\uff0c\u7ed3\u5408\u77e9\u9635\u5206\u89e3\u7684\u65b9\u6cd5\u663e\u8457\u63d0\u5347\u4e86\u5728\u771f\u5b9e\u4e16\u754c\u6570\u636e\u4e2d\u7684\u9884\u6d4b\u6027\u80fd\u3002\u6211\u4eec\u4ece\u7406\u8bba\u548c\u7ecf\u9a8c\u4e0a\u523b\u753b\u4e86\u8be5\u65b9\u6cd5\u7684\u9c81\u68d2\u6027\uff0c\u5e76\u4e14\u53d1\u73b0\u5b83\u5177\u6709\u9ad8\u5ea6\u7684\u53ef\u884c\u6027\u548c\u53ef\u6269\u5c55\u6027\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u9648\u6657\u66e6<\/span><\/section>\n

<\/section>\n
Chen, Minmin, Alex Beutel, Paul Covington, Sagar     Jain, Francois Belletti, and Ed H. Chi. “Top-k off-policy correction     for a REINFORCE recommender system.” In Proceedings of the Twelfth ACM International Conference on Web     Search and Data Mining, pp. 456-464. 2019.<\/span>
<\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u5de5\u4e1a\u754c\u63a8\u8350\u7cfb\u7edf\u4f1a\u5904\u7406\u975e\u5e38\u5927\u7684\u52a8\u4f5c\u7a7a\u95f4<\/span> (action spaces)<\/span>\u2014\u2014\u6570\u767e\u4e07 Items \u6765\u8fdb\u884c\u63a8\u8350\u3002\u540c\u65f6\uff0c\u4ed6\u4eec\u9700\u8981\u670d\u52a1\u6570\u5341\u4ebf\u7684\u7528\u6237\uff0c\u8fd9\u4e9b\u7528\u6237\u5728\u4efb\u610f\u65f6\u95f4\u70b9\u90fd\u662f\u552f\u4e00\u7684\uff0c\u4f7f\u5f97\u7528\u6237\u72b6\u6001\u7a7a\u95f4<\/span>\uff08user state space\uff09<\/span>\u5f88\u590d\u6742\u3002\u5e78\u8fd0\u7684\u662f\uff0c\u5b58\u5728\u6d77\u91cf\u7684\u9690\u5f0f\u53cd\u9988\u65e5\u5fd7\uff08\u6bd4\u5982\uff1a\u7528\u6237\u70b9\u51fb\uff0c\u505c\u7559\u65f6\u95f4\u7b49\uff09\u53ef\u7528\u4e8e\u5b66\u4e60\u3002\u4ece\u65e5\u5fd7\u53cd\u9988\u4e2d\u5b66\u4e60\u662f\u6709\u504f\u7684\uff0c\u8fd9\u662f\u56e0\u4e3a\u53ea\u6709\u5728\u63a8\u8350\u7cfb\u7edf\u4e0a\u89c2\u5bdf\u5230\u7684\u53cd\u9988\u662f\u7531\u4e4b\u524d\u7248\u672c\u7684\u63a8\u8350\u7cfb\u7edf\u9009\u4e2d\u7684\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u901a\u7528\u7684\u65b9\u6cd5\uff0c\u5728 Youtube \u751f\u4ea7\u73af\u5883\u4e0a\u7684 Top-k \u63a8\u8350\u7cfb\u7edf\u4e2d\uff0c\u4f7f\u7528\u4e00\u4e2a\u57fa\u4e8e\u7b56\u7565\u68af\u5ea6\u7684\u7b97\u6cd5<\/span>\uff08policy-gradient-based algorithm\uff0c\u6bd4\u5982\uff1aREINFORCE\uff09<\/span>\uff0c\u6765\u89e3\u51b3\u8fd9\u6837\u7684\u504f\u5dee\u3002\u8be5\u8bba\u6587\u7684\u4e3b\u8981\u8d21\u732e\u6709\uff1a1\uff09\u5c06 REINFORCE \u6269\u5c55\u5230\u751f\u4ea7\u73af\u5883\u63a8\u8350\u7cfb\u7edf\u4e0a\uff0c\u52a8\u4f5c\u7a7a\u95f4\u6709\u6570\u767e\u4e07\uff1b2\uff09\u4f7f\u7528 off-policy correction \u6765\u89e3\u51b3\u5728\u4ece\u591a\u79cd\u884c\u4e3a\u7b56\u7565\u4e2d\u6536\u96c6\u7684\u65e5\u5fd7\u53cd\u9988\u7684\u6570\u636e\u504f\u5dee\uff1b3\uff09\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684 Top-K off-policy correction \u6765\u89e3\u91ca\u6211\u4eec\u4e00\u6b21\u63a8\u8350\u591a\u4e2a Items \u7684\u7b56\u7565\u63a8\u8350\uff1b4\uff09\u5c55\u793a\u4e86\u63a2\u7d22\u7684\u4ef7\u503c\u6211\u4eec\u901a\u8fc7\u4e00\u7cfb\u5217\u4eff\u771f\u548c Youtube \u7684\u591a\u4e2a\u771f\u5b9e\u73af\u5883\uff0c\u6765\u5c55\u793a\u6211\u4eec\u7684\u65b9\u6cd5\u7684\u6548\u679c\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u9f9a\u9e64\u626c<\/span><\/section>\n

<\/section>\n
Yang, Longqi, Yin Cui, Yuan Xuan, Chenyang Wang, Serge Belongie, and Deborah Estrin. “Unbiased offline recommender evaluation for missing-not-at-random implicit feedback.” In Proceedings of the 12th ACM Conference on Recommender Systems, pp. 279-287. 2018.<\/span>
<\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u9690\u5f0f\u53cd\u9988\u63a8\u8350\u5668<\/span> (Implicit-feedback Recommenders, ImplicitRec)<\/span> \u53ea\u5229\u7528\u8bf8\u5982\u70b9\u51fb\u8fd9\u6837\u7684\u79ef\u6781\u7684\u201c\u7528\u6237-\u9879\u76ee\u201d\u4e92\u52a8\u6765\u5b66\u4e60\u4e2a\u6027\u5316\u7528\u6237\u504f\u597d\u3002\u4e0d\u540c\u63a8\u8350\u5668\u901a\u5e38\u4f7f\u7528\u4ece\u5728\u7ebf\u5e73\u53f0\u6536\u96c6\u7684\u6570\u636e\u96c6\u8fdb\u884c\u79bb\u7ebf\u8bc4\u4f30\u548c\u6bd4\u8f83\u3002\u8fd9\u4e9b\u5e73\u53f0\u53d7\u5230\u6d41\u884c\u5ea6\u504f\u5dee\u7684\u5f71\u54cd<\/span>\uff08\u6bd4\u5982\uff0c\u53d7\u6b22\u8fce\u7684\u9879\u76ee\u66f4\u5bb9\u6613\u88ab\u5448\u73b0\u7ed9\u7528\u6237\u5e76\u5f97\u5230\u4e92\u52a8\uff09<\/span>\uff0c\u56e0\u6b64\u65e5\u5fd7\u7684\u201c\u57fa\u51c6\u771f\u76f8\u6570\u636e\u201d<\/span>\uff08\u76d1\u7763\u5b66\u4e60\u7684\u6807\u6ce8\u6570\u636e\uff09<\/span>\u5b58\u5728\u5b8c\u5168\u975e\u968f\u673a\u7f3a\u5931<\/span> (Missing-Not-At-Random, MNAR)<\/span>\u3002\u6240\u4ee5\u88ab\u5e7f\u6cdb\u4f7f\u7528\u7684\u603b\u4f53\u5e73\u5747\u8bc4\u4f30\u5668<\/span> (Average-Over-All, AOA)<\/span> \u5728\u8bc4\u4f30\u53d7\u6b22\u8fce\u7684\u9879\u76ee\u7684\u51c6\u786e\u63a8\u8350\u65f6\u662f\u6709\u504f\u5dee\u7684\u3002\u5728\u8fd9\u7bc7\u6587\u7ae0\u4e2d\uff0c\u6211\u4eec\uff08a\uff09\u7814\u7a76\u4e86AOA\u7684\u8bc4\u4f30\u504f\u5dee\uff0c\u4ee5\u53ca\uff08b\uff09\u4f7f\u7528\u9006\u503e\u5411\u8bc4\u5206<\/span> (Inverse-Propensity-Scoring, IPS) <\/span>\u6280\u672f\u4e3a\u9690\u5f0fMNAR\u6570\u636e\u96c6\u5f00\u53d1\u4e00\u4e2a\u65e0\u504f\u4e14\u5b9e\u7528\u7684\u79bb\u7ebf\u8bc4\u4f30\u5668\u3002\u901a\u8fc7\u4f7f\u7528\u56db\u4e2a\u771f\u5b9e\u4e16\u754c\u6570\u636e\u96c6\u548c\u56db\u4e2a\u5e7f\u6cdb\u4f7f\u7528\u7684\u7b97\u6cd5\u7684\u5e7f\u6cdb\u5b9e\u9a8c\uff0c\u6211\u4eec\u8868\u660e\uff08a\uff09\u6d41\u884c\u5ea6\u504f\u5dee\u5e7f\u6cdb\u51fa\u73b0\u5728\u9879\u76ee\u5448\u73b0\u548c\u4e92\u52a8\u4e2d\uff1b\uff08b\uff09AOA\u7528\u4e8e\u8bc4\u4f30ImplicitRec\u65f6\uff0c\u7531\u4e8eMNAR\u6570\u636e\u5bfc\u81f4\u7684\u8bc4\u4f30\u504f\u5dee\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u666e\u904d\u5b58\u5728\uff1b\uff08c\uff09\u6839\u636e\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee(MAE)\uff0c\u65e0\u504f\u4f30\u8ba1\u663e\u8457\u51cf\u5c11\u4e86\u96c5\u864e\u97f3\u4e50\u6570\u636e\u96c6\u4e2d\u8d85\u8fc730%\u7684AOA\u8bc4\u4f30\u504f\u5dee\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u9648\u5929\u8c6a<\/span><\/section>\n

<\/section>\n
Bonner, Stephen, and Flavian Vasile. “Causal     embeddings for recommendation.” In Proceedings of the 12th ACM Conference on Recommender Systems,     pp. 104-112. 2018.<\/span><\/section>\n

<\/section>\n
\u7ffb\u8bd1\u6458\u8981\uff1a<\/span><\/strong><\/section>\n

\u8bb8\u591a\u73b0\u6709\u5e94\u7528\u4f7f\u7528\u63a8\u8350\u7684\u65b9\u5f0f\u6765\u4fee\u6539\u81ea\u7136\u7528\u6237\u7684\u884c\u4e3a\uff0c\u4f8b\u5982\u63d0\u5347\u9500\u91cf\u6216\u7f51\u7ad9\u7684\u6d4f\u89c8\u65f6\u95f4\u3002\u8fd9\u5bfc\u81f4\u4e86\u6700\u7ec8\u7684\u63a8\u8350\u76ee\u6807\u4e0e\u7ecf\u5178\u8bbe\u5b9a\u4ea7\u751f\u4e0d\u4e00\u81f4\uff0c\u5728\u7ecf\u5178\u8bbe\u7f6e\u4e2d\uff0c\u901a\u8fc7\u9884\u6d4buser-item matrix\u4e2d\u7f3a\u5c11\u7684\u6761\u76ee\u6216\u6700\u53ef\u80fd\u7684\u4e0b\u4e00\u4e2a\u4e8b\u4ef6\uff0c\u901a\u8fc7\u4e0e\u7528\u6237\u8fc7\u53bb\u884c\u4e3a\u7684\u4e00\u81f4\u6027\u6765\u8bc4\u4f30\u63a8\u8350\u5019\u9009\u3002\u4e3a\u4e86\u5f25\u8865\u8fd9\u79cd\u4e0d\u4e00\u81f4\uff0c\u6211\u4eec\u4f18\u5316\u4e86\u4e00\u4e2a\u7528\u4e8e\u76f8\u5bf9\u7528\u6237\u81ea\u7136\u884c\u4e3a\u589e\u52a0\u9884\u671f\u8f93\u51fa\u7684\u63a8\u8350\u7b56\u7565\u3002\u6211\u4eec\u5c55\u793a\u4e86\u8fd9\u4e2a\u7b56\u7565\u548c\u5728\u5b8c\u5168\u968f\u673a\u7684\u63a8\u8350\u7b56\u7565\u4e0b\u5b66\u4e60\u9884\u6d4b\u63a8\u8350\u8f93\u51fa\u662f\u7b49\u4ef7\u7684\u3002\u6700\u540e\uff0c\u6211\u4eec\u63d0\u51fa\u4e00\u79cd\u65b0\u7684\u9886\u57df\u81ea\u9002\u5e94\u7b97\u6cd5\uff0c\u5b83\u4ece\u5305\u542b\u504f\u5dee\u63a8\u8350\u7b56\u7565\u4ea7\u751f\u7684\u6570\u636e\u4e2d\u5b66\u4e60\uff0c\u9884\u6d4b\u968f\u673a\u63a8\u8350\u7b56\u7565\u4e0b\u7684\u63a8\u8350\u8f93\u51fa\u3002\u6211\u4eec\u5c06\u6211\u4eec\u7684\u65b9\u6cd5\u548c\u73b0\u6709\u7684SOTA\u5206\u89e3\u65b9\u6cd5\u8fdb\u884c\u6bd4\u8f83\uff0c\u5e76\u4e14\u548c\u65b0\u7684\u56e0\u679c\u63a8\u8350\u65b9\u6cd5\u6bd4\u8f83\uff0c\u7ed3\u679c\u53d6\u5f97\u4e86\u663e\u8457\u7684\u63d0\u5347\u3002<\/span><\/p>\n


<\/section>\n
\u8bd1\u8005\uff1a<\/span><\/strong>\u9648\u6657\u66e6<\/span><\/section>\n

<\/section>\n

<\/section>\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

2. \u4f1a\u8bae\uff1aRecSys2020\u7b49<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

RecSys 2020 <\/span>(14th ACM Conference on Recommender Systems Online, Worldwide, 22nd-26th September 2020) <\/span>\u662f\u63a8\u8350\u7cfb\u7edf\u65b9\u5411\u6700\u91cd\u8981\u7684\u4e00\u4e2a\u4f1a\u8bae\uff0c\u5b83\u5c55\u73b0\u4e86\u7684\u4e00\u4e2a\u91cd\u8981\u8d8b\u52bf\u662f\u5f3a\u8c03 B sias\uff0cRicardo Baeza-Yates \u505a\u4e86\u4e00\u4e2a\u5173\u4e8e\u6b64\u4e3b\u9898\u7684\u62a5\u544a\u6765\u7cfb\u7edf\u68b3\u7406\u5404\u79cd\u4e0d\u540c\u7684 bias \u6709\u5173\u7814\u7a76\u3002<\/span><\/p>\n


<\/span><\/section>\n
<\/section>\n
\u56fe3: \u63a8\u8350\u7cfb\u7edf\u4e2d\u7684 Bias<\/span><\/section>\n

<\/span><\/section>\n
\u9664\u4e86 RecSys \u4e4b\u5916\uff0c\u5176\u4ed6\u4e0e\u63a8\u8350\u7cfb\u7edf\u76f8\u5173\u7684\u91cd\u8981\u4f1a\u8bae\u5305\u62ec sigkdd, wsdm, sigir \u7b49\u3002<\/span><\/section>\n

<\/section>\n

<\/section>\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

3. \u8fd1\u671f\u793e\u533a\u6d3b\u52a8<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/section>\n
2020\u5e7411\u670815\u65e5\u4e0a\u53489\u70b9\uff0c\u201c\u56e0\u679c\u79d1\u5b66\u4e0eCausal AI\u201d\u8bfb\u4e66\u4f1a\u8fdb\u884c\u4e86\u7b2c8\u671f\u7684\u7ebf\u4e0a\u8bba\u6587\u5206\u4eab\uff0c\u4e3b\u9898\u662f\u201c\u56e0\u679c\u673a\u5668\u5b66\u4e60\u201d\u3002<\/span><\/section>\n

<\/section>\n
\u5206\u4eab\u8005\uff1a<\/span><\/strong>\u90ed\u82e5\u57ce\uff0c\u4e9a\u5229\u6851\u90a3\u5dde\u7acb\u5927\u5b66\u5728\u8bfb\u535a\u58eb<\/span><\/section>\n
\u4e2a\u4eba\u4e3b\u9875\uff1a<\/span><\/strong>https:\/\/www.public.asu.edu\/~rguo12<\/span>
<\/section>\n

<\/section>\n

\u8ba9\u6211\u4eec\u8003\u8651\u4e24\u79cd\u52a8\u7269\uff0c\u4f1a\u5b66\u4eba\u8bf4\u8bdd\u7684\u9e66\u9e49\u548c\u4e4c\u9e26\u559d\u6c34\u4e2d\u7684\u4e4c\u9e26\u3002\u6211\u4eec\u77e5\u9053\u9e66\u9e49\u53ea\u4f1a\u6a21\u4eff\u4eba\u7c7b\u7684\u8bed\u97f3\uff0c\u5374\u4e0d\u77e5\u9053\u8bed\u8a00\u7684\u610f\u4e49\u3002\u800c\u4e4c\u9e26\u53ef\u4ee5\u8ba4\u8bc6\u5230\u653e\u5165\u77f3\u5b50\u5c31\u53ef\u4ee5\u4f7f\u6c34\u9762\u4e0a\u5347\u8fd9\u4e00\u5c42\u56e0\u679c\u5173\u7cfb\u3002\u6211\u4eec\u53ef\u4ee5\u628a\u9e66\u9e49\u770b\u4f5c\u662f\u80fd\u505acurve fitting\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u800c\u628a\u4e4c\u9e26\u770b\u4f5c\u662f\u4f53\u73b0\u53d8\u91cf\u4e4b\u95f4\u56e0\u679c\u5173\u7cfb\u7684\u56e0\u679c\u6a21\u578b\u3002\u6bd4\u8d77\u95ee\u5b83\u4eec\u8c01\u66f4\u806a\u660e\uff0c\u6211\u60f3\u66f4\u6709\u610f\u4e49\u7684\u505a\u6cd5\u662f\u7ed3\u5408\u4ed6\u4fe9\u7684\u6280\u80fd\uff0c\u5e2e\u52a9\u6211\u4eec\u89e3\u51b3\u5b9e\u9645\u95ee\u9898\u3002\u5728\u673a\u5668\u5b66\u4e60\u5728\u5f88\u591a\u4efb\u52a1\u4e2d\u5df2\u53d6\u5f97\u6210\u529f\u7684\u4eca\u5929\uff0c\u6211\u4eec\u60f3\u56de\u7b54\u7684\u95ee\u9898\u662f\uff1a\u56e0\u679c\u63a8\u65ad\u80fd\u5426\u5e2e\u52a9\u673a\u5668\u5b66\u4e60\u5728\u4e00\u4e9b\u4efb\u52a1\u4e2d\u505a\u5f97\u66f4\u597d\uff0c\u4ee5\u53ca\u56e0\u679c\u63a8\u65ad\u80fd\u5426\u53d7\u76ca\u4e8e\u65b0\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\uff1f\u5728\u8fd9\u6b21\u7684\u8bfb\u4e66\u4f1a\u4e2d\uff0c\u6211\u4f1a\u4ece\u8fd9\u4e24\u4e2a\u65b9\u9762\u6765\u4ecb\u7ecd\u4e00\u4e9b\u73b0\u5b58\u7684\u7ed3\u5408\u56e0\u679c\u63a8\u65ad\u548c\u673a\u5668\u5b66\u4e60\u7684\u7814\u7a76\u65b9\u5411\u3002\u5728\u4f20\u7edf\u7684\u5229\u7528\u89c2\u6d4b\u6027\u6570\u636e\u7684\u56e0\u679c\u63a8\u65ad\u4e2d\uff0c\u5f88\u591a\u6a21\u578b\u5f80\u5f80\u901a\u8fc7\u5f88\u5f3a\u7684\u3001\u53ef\u80fd\u4e0d\u80fd\u88ab\u6ee1\u8db3\u7684\u5047\u8bbe\u6765\u907f\u514d\u8003\u8651hidden confounders\u3002\u800c\u6700\u8fd1\u7684\u4e00\u4e9b\u5de5\u4f5c\u4e2d\uff0c\u6211\u4eec\u53d1\u73b0\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u89c2\u6d4b\u6027\u6570\u636e\u4e2d\u9644\u5e26\u7684\u7f51\u7edc\u4fe1\u606f\u5f31\u5316\u8fd9\u4e00\u5047\u8bbe\uff0c\u63d0\u5347\u56e0\u679c\u63a8\u65ad\u6a21\u578b\u7684\u8868\u73b0\u3002\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u4e00\u4e9b\u95ee\u9898\u4e5f\u53ef\u4ee5\u5229\u7528\u5176\u80cc\u540e\u7684\u56e0\u679c\u6a21\u578b\u6765\u89e3\u51b3\u6216\u8005\u7f13\u89e3\u3002\u8fd9\u91cc\u7684\u7b2c\u4e00\u4e2a\u95ee\u9898\u662f\uff0c\u5982\u4f55\u4f7f\u795e\u7ecf\u7f51\u7edc\u66f4\u52a0\u9c81\u68d2\uff1f\u4ece\u56e0\u679c\u6027\u7684\u89d2\u5ea6\uff0c\u9c81\u68d2\u610f\u5473\u7740\u6211\u4eec\u8981\u907f\u514d\u5b66\u4e60\u5230spurious correlation\uff1f\u6bd4\u5982\u5728\u52a8\u7269\u56fe\u50cf\u5206\u7c7b\u4e2d\uff0c\u9a86\u9a7c\u5e38\u5e38\u51fa\u73b0\u5728\u6c99\u6f20\u4e2d\uff0c\u5982\u4f55\u80fd\u907f\u514d\u5206\u7c7b\u5668\u5229\u7528\u6c99\u6f20\u80cc\u666f\u6765\u9884\u6d4b\u9a86\u9a7c\u8fd9\u4e2a\u7c7b\uff1f\u6700\u8fd1\uff0c\u5728Invariant Risk Minimization (IRM) \u8fd9\u4e2a\u5de5\u4f5c\u4e2d\uff0c\u4f5c\u8005\u4eec\u628a\u5177\u4f53\u95ee\u9898\u4e2d\u56e0\u679c\u5173\u7cfb\u7684\u9650\u5236\u6761\u4ef6\u8f6c\u5316\u6210\u4e86\u673a\u5668\u5b66\u4e60\u6a21\u578b\u4e2d\u7684inductive bias\uff0c\u4ece\u800c\u8fbe\u5230\u8fd9\u4e00\u76ee\u7684\u3002\u800c\u5728\u6211\u4eec\u7684\u5de5\u4f5c\u4e2d\uff0c\u6211\u4eec\u53d1\u73b0IRM\u662f\u4e00\u4e2a\u8fc7\u4e8e\u5bbd\u6cdb\u7684\u6761\u4ef6\u3002\u9488\u5bf9\u8fd9\u4e00\u95ee\u9898\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u7b80\u5355\u6709\u6548\u7684\u89e3\u51b3\u65b9\u6848\u3002\u7b2c\u4e8c\u4e2a\u95ee\u9898\u662f\uff0c\u5728\u4f7f\u7528\u7528\u6237\u53cd\u9988\u4f5c\u4e3a\u6807\u7b7e\u7684\u673a\u5668\u5b66\u4e60\u4efb\u52a1\u4e2d\uff0c\u5982\u4f55\u505a\u5230\u5229\u7528\u5386\u53f2\u6570\u636e\u5728\u7ebf\u4e0b\u5bf9\u65b0\u6a21\u578b\u8fdb\u884c\u8bc4\u6d4b\u548c\u4f18\u5316\uff0c\u4f7f\u5176\u7ebf\u4e0a\u6548\u679c\u5f97\u5230\u63d0\u5347\u3002\u5728\u6700\u8fd1\u7684\u63a8\u8350\u7cfb\u7edf\u548c\u641c\u7d22\u7684\u5de5\u4f5c\u4e2d\uff0c\u4eba\u4eec\u53d1\u73b0\u628a\u7528\u6237\u53cd\u9988\u6807\u7b7e\u7684\u56e0\u679c\u56fe\u878d\u5165\u6a21\u578b\u8bbe\u8ba1\u5c06\u4f7f\u6211\u4eec\u5728\u8fd9\u4e00\u4efb\u52a1\u4e2d\u505a\u5f97\u66f4\u597d\u3002<\/span><\/p>\n


<\/span><\/section>\n
\u6f14\u8bb2\u5927\u7eb2\uff1a<\/span><\/strong><\/span><\/section>\n
Machine learning for causal inference<\/span><\/section>\n
 * Learning causality with networked observational data<\/span><\/section>\n
Causality-aware machine learning<\/span><\/section>\n
 * Out-of-distribution prediction with causal inductive bias<\/span><\/section>\n
 * Unbiased interactive machine learning<\/span><\/section>\n

<\/span><\/section>\n

\u5728\u673a\u5668\u5b66\u4e60\u5df2\u7ecf\u5728\u5f88\u591a\u4efb\u52a1\u4e2d\u5df2\u7ecf\u53d6\u5f97\u6210\u529f\u7684\u4eca\u5929\uff0c\u6211\u4eec\u60f3\u56de\u7b54\u7684\u95ee\u9898\u662f\uff1a\u56e0\u679c\u63a8\u65ad\u80fd\u5426\u5e2e\u52a9\u673a\u5668\u5b66\u4e60\u5728\u4e00\u4e9b\u4efb\u52a1\u4e2d\u505a\u5f97\u66f4\u597d\u4ee5\u53ca\uff0c\u56e0\u679c\u63a8\u65ad\u80fd\u5426\u53d7\u76ca\u4e8e\u65b0\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\uff1f\u5728\u8fd9\u6b21\u7684\u8bfb\u4e66\u4f1a\u4e2d\uff0c\u90ed\u82e5\u57ce\u5c06\u4f1a\u4ece\u8fd9\u4e24\u4e2a\u65b9\u9762\u6765\u4ecb\u7ecd\u4e00\u4e9b\u73b0\u5b58\u7684\u7ed3\u5408\u56e0\u679c\u63a8\u65ad\u548c\u673a\u5668\u5b66\u4e60\u7684\u7814\u7a76\u65b9\u5411\u3002\u5728\u4f20\u7edf\u7684\u5229\u7528\u89c2\u6d4b\u6027\u6570\u636e\u7684\u56e0\u679c\u63a8\u65ad\u4e2d\uff0c\u5f88\u591a\u6a21\u578b\u5f80\u5f80\u901a\u8fc7\u5f88\u5f3a\u7684\uff0c\u53ef\u80fd\u4e0d\u80fd\u88ab\u6ee1\u8db3\u7684\u5047\u8bbe\u6765\u907f\u514d\u8003\u8651hidden confounders\u3002\u800c\u6700\u8fd1\u7684\u4e00\u4e9b\u5de5\u4f5c\u4e2d\uff0c\u4ed6\u4eec\u53d1\u73b0\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u53ef\u4ee5\u5229\u7528\u89c2\u6d4b\u6027\u6570\u636e\u4e2d\u9644\u5e26\u7684\u7f51\u7edc\u4fe1\u606f\u5f31\u5316\u8fd9\u4e00\u5047\u8bbe\uff0c\u63d0\u5347\u56e0\u679c\u63a8\u65ad\u6a21\u578b\u7684\u8868\u73b0\u3002<\/span><\/p>\n


<\/section>\n
\u8bfb\u4e66\u4f1a\u7cbe\u5f69\u95ee\u7b54\uff1a<\/span><\/strong><\/section>\n

<\/span><\/section>\n
\u6574\u7406\uff1a<\/span>\u6768\u4e8c\u8336<\/span>
<\/section>\n

<\/section>\n

Q\uff1a\u8bf7\u95ee\u4e00\u4e0b\u82e5\u57ce\uff0c\u662f\u5426\u5173\u6ce8\u8fc7\u56e0\u679c\u63a8\u65ad\u548c\u63a8\u8350\u7cfb\u7edfmatch\u7684\u5de5\u4f5c\uff1f\u4ee5\u53ca\u8fd9\u4e9b\u5de5\u4f5c\u662f\u5426\u6d89\u53cacofounding\u7684\u95ee\u9898\uff1f<\/span><\/strong><\/p>\n


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

A\uff1a<\/strong>\u7528\u63a8\u8350\u7cfb\u7edf\u7684\u79bb\u7ebf\u6570\u636e\u53bbmatch\u5728\u7ebfdistribution\u66fe\u7ecf\u6709\u4e00\u7bc7\u9876\u4f1a\u6587\u7ae0\u53ebCausal embedding accommodation\u7528\u8fc7\u8fd9\u4e2a\u601d\u60f3\u3002\u673a\u5668\u5b66\u4e60\u4e2d\u63d0confounding\u5176\u5b9e\u662f\u5f88\u5947\u602a\u7684\uff0c\u6211\u4eec\u77e5\u9053\uff0cconfounding\u5bfc\u81f4bias\u7684\u539f\u56e0\u662f\u56e0\u4e3acausal effect estimation\u3002\u6240\u4ee5\u5f53\u4f60\u9700\u8981\u6d4b\u5ea6T->Y\u7684\u56e0\u679c\u6548\u5e94\uff0c\u4f60\u53bb\u5efa\u7acbconfounder\u7684\u5b9e\u8d28\u662f\u4e3a\u4e86\u5efa\u7acb\u5176\u4ed6\u53d8\u91cf\u7684\u5173\u7cfb\uff0c\u56e0\u6b64\uff0c\u4f60\u5fc5\u987b\u5f97\u6709\u6570\u636e\u5206\u5e03\u4e0a\u7684\u53d8\u5316\uff0c\u624d\u80fd\u53bb\u89e3\u51b3confounding\u7684\u95ee\u9898\u3002\u673a\u5668\u5b66\u4e60\u5f88\u5c11\u53bb\u63a2\u7a76\u4e00\u4e2a\u53d8\u91cfT\u5230\u53e6\u4e00\u4e2a\u53d8\u91cfY\u7684\u5f71\u54cd\uff0c\u4f60\u53ea\u60f3\u9884\u6d4bY\u800c\u5df2\u3002\u63a8\u8350\u7cfb\u7edf\u4e2d\u4e5f\u6709\u5f88\u591a\u7814\u7a76\uff0c\u63a8\u8350\u7cfb\u7edf\u5bf9population\u6709\u4ec0\u4e48\u56e0\u679c\u6548\u5e94\uff0c\u8fd9\u79cd\u60c5\u51b5\u4e5f\u662f\u53ef\u4ee5\u901a\u8fc7condition on X\u6765\u89e3\u51b3\u3002<\/span>
<\/p>\n


<\/p>\n

Q\uff1a\u6709\u6ca1\u6709\u4e00\u4e9b\u6587\u7ae0\u6765\u89e3\u51b3bias\u7684\u95ee\u9898\uff1f\u4e3e\u4e2a\u4f8b\u5b50\uff0c\u5982\u679c\u8fd9\u4e2a\u573a\u666f\u662f\u7ed9\u75c5\u4eba\u5403\u836f\uff0c\u4f46\u836f\u6709\u4e0d\u540c\u6210\u5206\uff0c\u4e0d\u540c\u75c5\u4eba\u7684\u6210\u5206\u7ec4\u5408\u4e0d\u540c\uff0c\u5bf9\u75c5\u4eba\u62bd\u6837\u6709bias\uff0c\u90a3\u4e48\u6211\u4eec\u5982\u679c\u8fdb\u884cdebias\u7684\u5de5\u4f5c\u5462\uff1f\u6211\u4eec\u4e3b\u89c2\u4e0a0\u7684\u5206\u7c7b\u5e76\u4e0d\u4e00\u5b9a\u662f\u771f\u5b9e\u573a\u666f\u4e2d\u7684\u5206\u7c7b\uff1f\u5982\u679c\u6211\u627e\u5230\u4e00\u4e2aconfounder\uff0c\u4f46\u540c\u65f6confounder\u53c8\u6709\u5f88\u591a\u7684\u8bdd\uff0c\u5e94\u8be5\u5982\u4f55\u5904\u7406\uff1f<\/span><\/strong><\/p>\n


<\/p>\n

A\uff1a<\/span><\/strong>causal inference\u7684\u6838\u5fc3\u89e3\u51b3\u7684\u5c31\u662f\u8fd9\u4e2a\u95ee\u9898\uff0c\u901a\u8fc7control confounder\u6765\u89e3\u51b3miss\u7684confounder\u6765\u5efa\u6a21\u3002\u6240\u4ee5\u5176\u672c\u8d28\u662f\u4e00\u4e2aunobserved cofounder\u7684\u95ee\u9898\uff0c\u56e0\u4e3a\u4f60\u4e0d\u53ef\u80fd\u77e5\u9053\u6240\u6709\u7684confounder\uff0c\u6240\u4ee5\u4f60\u53ea\u80fd\u5c3d\u91cf\u53d1\u73b0\u66f4\u591a\u7684confounder\u6765\u9884\u6d4btreatment effect\u3002\u4e0d\u63a7\u5236confounder\u7684\u8bdd\uff0c\u5c31\u5f88\u96be\u53bb\u6389\u8fd9\u79cd\u5e72\u6270\u7684\u5f71\u54cd\u3002\u5176\u5b9e\u73b0\u5728\uff0c\u9664\u4e86\u968f\u673a\u5b9e\u9a8c\u4ee5\u5916\uff0c\u6211\u4eec\u8fd8\u6ca1\u6709\u5176\u4ed6\u529e\u6cd5\u53ef\u4ee5\u53bb\u6389\u5176\u4ed6confounder<\/span>\uff08\u5bf9\u4e3b\u6548\u5e94\uff09<\/span>\u7684\u5f71\u54cd\u3002<\/span><\/p>\n


<\/section>\n


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

\n
\n
\n
\n

\u7cfb\u5217\u8bfb\u4e66\u4f1a\u62a5\u540d<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u5927\u6570\u636e\u65f6\u4ee3\u7684\u4e0b\u4e00\u573a\u53d8\u9769\u2014\u2014\u56e0\u679c\u9769\u547d\u6b63\u5728\u915d\u917f\u4e4b\u4e2d\uff0c\u901a\u8fc7\u878d\u5408\u56e0\u679c\u63a8\u7406\u548c\u673a\u5668\u5b66\u4e60\u800c\u6784\u5efa\u51fa\u6765\u7684Causal AI\u7cfb\u7edf\uff0c\u6709\u671b\u5960\u5b9a\u5f3a\u4eba\u5de5\u667a\u80fd\u7684\u57fa\u77f3\u3002<\/span><\/p>\n


<\/p>\n

\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u5317\u4eac\u667a\u6e90\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u9662\uff0c\u9080\u8bf7\u4e86\u4e00\u6279\u5bf9\u56e0\u679c\u79d1\u5b66\u4e0eCausal AI\u611f\u5174\u8da3\u7684\u7814\u7a76\u8005\uff0c\u5f00\u5c55\u4e3a\u671f2-3\u4e2a\u6708\u7684\u7cfb\u5217\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\uff0c\u7814\u8bfb\u7ecf\u5178\u548c\u524d\u6cbf\u8bba\u6587\uff0c\u5e76\u5c1d\u8bd5\u96c6\u4f53\u64b0\u5199\u4e00\u90e8\u4e66\u7c4d\u3002\u5982\u679c\u4f60\u4e5f\u4ece\u4e8b\u76f8\u5173\u7684\u7814\u7a76\u3001\u5e94\u7528\u5de5\u4f5c\uff0c\u6b22\u8fce\u62a5\u540d\uff0c\u53c2\u4e0e\u8bfb\u4e66\u4f1a\u7684\u8ba8\u8bba\uff01<\/span><\/p>\n

 <\/span><\/p>\n

\u65f6\u95f4\uff1a<\/span><\/strong>9\u670820\u65e5\u8d77\uff0c\u6bcf\u5468\u65e5\u665a19:00-21:00\uff0c\u6301\u7eed\u7ea62-3\u4e2a\u6708<\/span><\/section>\n
\u6a21\u5f0f\uff1a<\/span><\/strong>\u7ebf\u4e0a\u95ed\u95e8\u8bfb\u4e66\u4f1a\uff1b\u6536\u8d39-\u9000\u6b3e\u7684\u4fdd\u8bc1\u91d1\u6a21\u5f0f\uff1b\u8bfb\u4e66\u4f1a\u6210\u5458\u8ba4\u9886\u89e3\u8bfb\u8bba\u6587<\/span><\/section>\n
\u8d39\u7528\uff1a<\/span><\/strong>299\/\u4eba<\/span><\/section>\n

 <\/span><\/p>\n

\u4e86\u89e3\u8bfb\u4e66\u4f1a\u5177\u4f53\u89c4\u5219\u3001\u62a5\u540d\u8bfb\u4e66\u4f1a\u8bf7\u70b9\u51fb\u4e0b\u65b9\u6587\u7ae0\uff1a<\/span><\/p>\n

\u56e0\u679c\u79d1\u5b66\u4e0e Causal AI \u7cfb\u5217\u8bfb\u4e66\u4f1a | \u4f17\u5305\u51fa\u4e66<\/a><\/section>\n


<\/p>\n

\u76ee\u524d\u8bfb\u4e66\u4f1a\u5df2\u7ecf\u6709\u8d85\u8fc7130\u4f59\u4eba\u7684\u6d77\u5185\u5916\u9ad8\u6821\u79d1\u7814\u9662\u6240\u7684\u4e00\u7ebf\u79d1\u7814\u5de5\u4f5c\u8005\u4ee5\u53ca\u4e92\u8054\u7f51\u4e00\u7ebf\u4ece\u4e1a\u4eba\u5458\u53c2\u4e0e\uff0c\u5982\u679c\u4f60\u4e5f\u5bf9\u8fd9\u4e2a\u4e3b\u9898\u611f\u5174\u8da3\uff0c\u5c31\u5feb\u52a0\u5165\u6211\u4eec\u5427\uff01<\/span><\/p>\n

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

              <\/span><\/p>\n

\u56fe\u6ce8\uff1a\u9488\u5bf9\u8bfb\u4e66\u4f1a\u7684\u4e3b\u9898\uff0c\u7531\u53d1\u8d77\u4eba\u9f9a\u9e64\u626c\u8bbe\u7f6e\u597d\u4e86\u5185\u5bb9\u6846\u67b6\uff0c\u6bcf\u4e2a\u4e3b\u9898\u4e0b\u6709\u4e00\u4e2a\u8d1f\u8d23\u4eba\u6765\u8d1f\u8d23\u7ef4\u62a4\u7ec4\u7ec7\u76f8\u5173\u5185\u5bb9\uff0c\u76ee\u524d\u5df2\u7ecf\u5b9a\u597d\u7684\u5982\u56fe\u6240\u793a\uff0c\u6b22\u8fce\u5bf9\u4e3b\u9898\u611f\u5174\u8da3\u7684\u8054\u7cfb\u76f8\u5173\u8d1f\u8d23\u4eba\uff0c\u4ee5\u53ca\u6765\u8ba4\u9886\u76f8\u5173\u4e3b\u9898\u3002<\/em><\/span><\/p>\n<\/p><\/div>\n","protected":false},"excerpt":{"rendered":"

\u4e3a\u4e86\u5e2e\u52a9\u5927\u5bb6\u66f4\u597d\u5730\u4e86\u89e3\u56e0\u679c\u79d1\u5b66\u7684\u6700\u65b0\u79d1\u7814\u8fdb\u5c55\u548c\u8d44\u8baf\uff0c\u6211\u4eec\u56e0\u679c\u79d1\u5b66\u793e\u533a\u56e2\u961f\u672c\u5468\u6574\u7406\u4e86\u7b2c4\u671f\u300a\u56e0\u679c\u79d1\u5b66\u5468\u520a\u300b\uff0c\u63a8\u9001\u8fd1\u671f\u56e0\u679c\u79d1\u5b66\u503c\u5f97\u5173\u6ce8\u7684\u8bba\u6587\u548c\u8d44\u8baf\u4fe1\u606f\uff0c \u540c\u65f6\u6211\u4eec\u4e5f\u5c06\u5411\u5927\u5bb6\u4ecb\u7ecd\u793e\u533a\u6b63\u5728\u63a8\u8fdb\u7684\u6d3b\u52a8\u2014\u2014\u56e0\u679c\u79d1\u5b66\u4e0eCausal AI\u8bfb\u4e66\u4f1a\u7b2c8\u671f\u4e2d\u7684\u4e3b\u8981\u62a5\u544a\u5185\u5bb9\u3001\u89c2\u70b9\u3002 \u672c\u671f\u4f5c\u8005\uff1a\u4faf\u8339 \u95eb\u548c\u4e1c \u9648\u6657\u66e6 \u9f9a\u9e64\u626c …<\/p>\n","protected":false},"author":1,"featured_media":22354,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/22362"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=22362"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/22362\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/22354"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=22362"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=22362"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=22362"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=22362"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}