英语翻译To illustrate our learning model,consider the example of a personalized news reader that users visit on a daily basis.On day t,the news reader suggests a list of articles yt = (d1,d2,d3,d4,d5,...) and observes which of these articles are

英语翻译To illustrate our learning model,consider the example of a personalized news reader that users visit on a daily basis.On day t,the news reader suggests a list of articles yt = (d1,d2,d3,d4,d5,...) and observes which of these articles are
英语翻译
To illustrate our learning model,consider the example of a personalized news reader that users visit on a daily basis.On day t,the news reader suggests a list of articles yt = (d1,d2,d3,d4,d5,...) and observes which of these articles are actually read by the user.We assume that the decision to read an article is influenced by two factors.First,the article must be relevant to the user’s interest.Second,the decision may have dependencies with other articles in y.For example,the user may be interested in the European debt crisis.But the user may only want to read one article related to this issue,even if y contains 5 relevant articles.
In this paper,we design an online learning algorithm that can model both relevance as well as interdependencies be- tween documents.The training data we exploit are the sets of documents read by the user each day.Continu- ing the example from above,the system may observe that the user read articles d3 and d5.Obviously,we cannot conclude that {d3,d5} was the optimal set of articles the user wanted to read on day t,since there may have been other articles far down the list that the user never saw.However,we can conclude that the user would have pre- ferred the ranking y¯t = (d3,d5,d1,d2,d4,...) over the rank- ing yt = (d1,d2,d3,d4,d5,...) that was presented.We refer to y¯t as the user feedback ranking.

英语翻译To illustrate our learning model,consider the example of a personalized news reader that users visit on a daily basis.On day t,the news reader suggests a list of articles yt = (d1,d2,d3,d4,d5,...) and observes which of these articles are
为描述我们的学习模型,让我们想这样一个例子,一个用户每天都会访问的个性化新闻阅读器.在第t天,新闻阅读器推荐了一系列文章 yt = (d1,d2,d3,d4,d5,...) ,并且观测哪篇文章真的被用户阅读过了.我们假设决定阅读那篇文章是受两个因素影响的.其一,文章必须和用户的兴趣相关.其二,这个决定可能会依赖于y里面的其它文章.比如说,用户可能会对欧洲债务危机感兴趣,但是就算y里面包括了5篇相关文章,他大概只会阅读与此问题相关的一篇文章.
在本文中,我们设计了一个在线的学习算法,它能够同时以相关性和文件之间的相互依赖性来建模.我们采用的训练数据是用户每天阅读的系列文件.继续用上面的例子,该系统可能会观测到用户阅读了d3和d5.显而易见的,我们不能下结论说 {d3,d5}是用户想在第t天阅读的最佳文章集合,因为列表的下部可能还有用户根本就没瞧见的别的文章.然而我们可以得出下面的结论,用户对于出示的列表yt = (d1,d2,d3,d4,d5,...) ,会有个优选排行 y¯t = (d3,d5,d1,d2,d4,...).我们把y¯t = (d3,d5,d1,d2,d4,...)称作用户反馈排行.

为了说明我们的学习模型,考虑的例子,一个个性化的新闻阅读器,用户每天的访问。在天t,新闻阅读器显示一个物品清单刘日东=(d1、d2、d3、d4、d5、…)和观察,这些文章都是由用户实际阅读。我们假设决定读了一篇文章是由两个因素的影响。首先,文章必须与用户相关的利益。第二,这一决定可能会依赖与其他文章在y。例如,用户可能感兴趣的欧洲债务危机。但用户可能只想读一篇与此相关的问题,即使y包含5相关文章。...

全部展开

为了说明我们的学习模型,考虑的例子,一个个性化的新闻阅读器,用户每天的访问。在天t,新闻阅读器显示一个物品清单刘日东=(d1、d2、d3、d4、d5、…)和观察,这些文章都是由用户实际阅读。我们假设决定读了一篇文章是由两个因素的影响。首先,文章必须与用户相关的利益。第二,这一决定可能会依赖与其他文章在y。例如,用户可能感兴趣的欧洲债务危机。但用户可能只想读一篇与此相关的问题,即使y包含5相关文章。
在本文中,我们设计一个在线学习算法,该算法能模型两个相关性以及相互依赖的是——二层文件。训练数据我们利用是套文件读取用户每一天。Continu - ing上面的示例,系统可能会观察到用户阅读文章和d5 d3。显然,我们不能得出{ d3,d5 }是最优组的文章用户想看天,因为可能会有不被其他文章深列表,用户从未见过。然而,我们可以得出结论,用户会pre -转让排名y¯t =(d3,d5、d1、d2 d4,…)- ing刘日东排名=(d1、d2、d3、d4、d5,…),给出了。我们称y¯t作为用户反馈的排名。

收起