{"resolved":true,"match_type":"uuid","source_type":"original_book","book_id":"06dd97ec-cce5-45df-9acf-40e36f07dab3","title":"[数字经济前沿]技术封建主义","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787300328508","requested_by":"printed_page","book":{"source_type":"original_book","book_id":"06dd97ec-cce5-45df-9acf-40e36f07dab3","id":"06dd97ec-cce5-45df-9acf-40e36f07dab3","uuid":"06dd97ec-cce5-45df-9acf-40e36f07dab3","title":"[数字经济前沿]技术封建主义","normalized_title":"数字经济前沿技术封建主义","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787300328508","isbns":["9787300328508"],"parsed":true,"ai_readable":true,"document_type":"book","page_count":259,"navigation_url":"https://workshop.closeclaw.top/ai/book/06dd97ec-cce5-45df-9acf-40e36f07dab3","title_url":"https://workshop.closeclaw.top/ai/title/%5B%E6%95%B0%E5%AD%97%E7%BB%8F%E6%B5%8E%E5%89%8D%E6%B2%BF%5D%E6%8A%80%E6%9C%AF%E5%B0%81%E5%BB%BA%E4%B8%BB%E4%B9%89","fixed_page_url_template":"https://workshop.closeclaw.top/ai/book/06dd97ec-cce5-45df-9acf-40e36f07dab3/page/{printed_page}","path_api_url_template":"https://workshop.closeclaw.top/api/ai/book/06dd97ec-cce5-45df-9acf-40e36f07dab3/page/{printed_page}","read_url":"https://workshop.closeclaw.top/api/ai/read/06dd97ec-cce5-45df-9acf-40e36f07dab3","page_url":"https://workshop.closeclaw.top/api/ai/books/06dd97ec-cce5-45df-9acf-40e36f07dab3/page","search_url":"https://workshop.closeclaw.top/api/ai/books/06dd97ec-cce5-45df-9acf-40e36f07dab3/search"},"page":{"printed_page":101,"pdf_page":141,"physical_index":141,"physical_index_base":0,"printed_page_status":"recognized","page_sha256":"67a67dc42ee996b6b5b2926a300457a2f633fab3d341ffd40de1da53016bf65a"},"chapter":"第二章","content":"随着互联网的发展，在线对照实验从20世纪90年代末开始使用。如今，许多大型网站，包括亚马逊、必应（Bing)、脸书、谷歌、领英（LinkedIn）和雅虎每年都会进行成千上万次实验，以测试用户界面的变化、算法（搜索、广告、个性化、推荐等）的改进、应用程序的变更、内容管理系统等。如今，在线实验被视作不可或缺的工具，并且在初创企业和小型网站中的使用日益增多。①\n\n这种大规模应用的原因很简单，联网对象的激增使得大规模评估不同设备对个人行为的影响成为可能。与其他数据挖掘技术不同的是，这种技术的目的不是简单地显示自发的统计规律，而是以极高的概率建立可控的相关性。\n\n实验者通常是经济学家②，他们可以探索以下假设：“如果引入特定的变化，是否会改善关键指标？”通常，实验可能稍微修改搜索引擎上广告的边缘形状，然后将标准视图和修改后的视图随机分发给用户，以观察用户行为的变化。例如，在微软搜索引擎必应上进行的一项测试中，在广告横幅中添加内部链接，或改变字体的大小，甚至改变页面上的广告数量。引入的差异往往很小，但它们可能产生价值数百万美元的影响。\n\n第二章 数字统治\n\n① Ron KOHAVI et Roger LONGBOTHAM, «Online controlled experiments and A/B testing», in Claude SAMMUT et Geoffrey I. WEBB (dir.), Encyclopedia of Machine Learning and Data Mining, Springer, Boston, 2017, p. 922.\n\n② Susan ATHEY et Michael LUCA, «Economists (and economics) in tech companies», NBER Working Paper, n° 19-027, 2018, p. 7.\n\n101","previous_page":100,"next_page":102}