人工智能原理人工智能原理 (57).pdf

《人工智能原理人工智能原理 (57).pdf》由会员分享，可在线阅读，更多相关《人工智能原理人工智能原理 (57).pdf（12页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、Artificial Intelligence68 10.1.Classification 10.2.Regression 10.3.Clustering 10.4.Ranking 10.5.Dimensionality ReductionContents:10.Tasks in Machine LearningRankingArtificial Intelligence:Learning:Tasks70 A longer description 较长描述A ranking is a relationship between a set of items such that,for any t

2、wo items,the first is either ranked higher than,ranked lower than or ranked equal to the second.排名是一组项之间的关系，即对于任意两个项，满足第一个“排名高于”、“排名低于”或“排名等于”第二个。A shorter description 较短描述The data transformation in which numerical or ordinal values are replaced by their rank.排名是一种数据转换，其中数值或者顺序值由其排名来代替。A very short

3、description 极简描述To order items according to some criterion.依据某种准则整理数据项。What is Ranking 什么是排名10.4.RankingArtificial Intelligence71 10.4.1.How Ranking Works 10.4.2.Major Approaches of Ranking 10.4.3.Applications and AlgorithmsContents:10.4.RankingArtificial Intelligence:Learning:Tasks72Let X denote in

4、put space,D an unknown distribution over X X.设X表示输入空间，D是X X 上的未知分布。Target ranking function:目标排名函数：A Formal Description of Ranking 一种排名的形式化描述10.4.1.How Ranking Worksf X X Y=1,0,+1S=(x(i),x(i),y(j)|y(j)=f(x(i),x(i)Y,i 1,m,j 1,3 Training data:训练数据where f(x,x)=+1,if x is ranked higher than x,f(x,x)=1,if

5、 x is ranked lower than x,f(x,x)=0,if both x and x has same ranking.其中若x排名高于x，若x排名低于x，若x与x二者排名相同。Artificial Intelligence:Learning:Tasks73 Ranking problem:排名问题Given a hypothesis set H of functions mapping X X to Y=1,0,+1,to select a hypothesis h H with the target function f:给定一个将X X 映射到Y=1,0,+1的假设函数集

6、H，选择一个具有目标函数 f 的假设h H：small expected generalization error:最小预期泛化错误：A Formal Description of Ranking 一种排名的形式化描述10.4.1.How Ranking Works(h)=1=11()0)()(h()h()0)R(h)=Pr(x,x)f(x,x)0 (f(x,x)(h(x)h(x)0)empirical pairwise misranking error:经验性成对误排名错误：Artificial Intelligence74 10.4.1.How Ranking Works 10.4.2.M

7、ajor Approaches of Ranking 10.4.3.Applications and AlgorithmsContents:10.4.RankingArtificial Intelligence:Learning:Tasks75 1)Score-based approach 基于分值方法 The predictor is a real-valued function,called scoring function.该预测器是一个实数函数，称为分值函数。The scores assigned to input points by this function determine t

8、heir ranking.由该函数分派给输入数据点的分值决定其排名。This approach is the most widely explored one.这种方法是研究得最多的一种。2)Preference-based approach 基于偏好方法 The predictor is a preference function.该预测器是一个偏好函数。Typical Approaches of Ranking 典型的排名方法10.4.2.Major Approaches of RankingArtificial Intelligence76 10.4.1.How Ranking Work

9、s 10.4.2.Major Approaches of Ranking 10.4.3.Applications and AlgorithmsContents:10.4.RankingArtificial Intelligence:Learning:Tasks77 In information retrieval Search engine Document retrieval Collaborative filtering Sentiment analysis Computational advertising In other areas Machine translation Recom

10、mender systems Computational biology ProteomicsTypical Applications of Ranking 排名的典型应用10.4.3.Applications and Algorithms信息检索领域搜索引擎文档检索协同式过滤情感分析计算广告学其它领域机器翻译推荐系统计算生物学蛋白质组学Artificial Intelligence:Learning:Tasks78 An algorithm used by Google to rank websites in their search engine,named after Larry Pag

11、e,one of Google founders.谷歌用于在其搜索引擎中对网站进行排名的一种算法，以谷歌创始人之一拉里佩奇的名字命名。Case Study:PageRank10.4.3.Applications and Algorithms PageRank works by counting the number and quality of links to a page to determine how important the website is.PageRank通过计算网页的链接数量和质量来决定该网站的重要性。The underlying assumption is that more important websites are likely to receive more links from other websites.其基本假设是：越重要的网站，就会被越多其它网站所链接。