Presentation_of_ListMLE_读书会第8期.pptVIP

* * * From 1898 to 1978 to 2008 Browse is impossible Search engine – essential tool 信息按一定的方式组织起来，并根据信息用户的需要找出有关的信息的过程和技术。狭义的信息检索就是信息检索过程的后半部分，即从信息集合中找出所需要的信息的过程，也就是我们常说的信息查寻 * 查询和匹配的语境和要求不同: 信息检索类： 特征描述丰富， 相对排序准确，top-k准确性； 协作过滤： 特征描述不丰富，绝对排序或等级； 序回归： 特征描述丰富，绝对排序，介于分类和回归之间； * * * * * * 机器学习： 目标函数+求解算法 * * * More: surrogate loss can be top-1 subgroup order sensitive to meet the consistency condition. * * * * As for the likelihood loss, the situation becomes much better. When the green point moves from the lower right to the upper left of the space, the likelihood loss decreases monotonously, and the minimum loss is achieved at the infinity of the upper left part of the space. This is clearly more reasonable than the aforementioned two losses. So overall speaking, by jointly considering all the properties, the likelihood loss is better than the cosine and the cross entropy loss. This partially explains why ListMLE outperforms other ranking algorithms. * * Learning to Rank – Theory and Algorithm @夏粉_百度 合办方：超级计算大脑研究部@自动化所 * We are Overwhelmed by Flood of Information * Information Explosion * 2013? * Ranking Plays Key Role in Many Applications * Numerous Applications Ranking Problem Information Retrieval Collaborative Filtering Ordinal Regression Example Applications * Overview of my Work before 2010 Machine Learning Theory and Principle Ranking Problems Information Retrieval Collaborative Filtering Ordinal Regression Theory Algorithm NIPS’09 PR’09 ICML’08 JCST’09 KAIS’08 IJICS’07 IJCNN’07 IEEE-IIB’06 * Outline Listwise Approach to Learning to Rank – Theory and Algorithm Related Work Our Work Future Work * Ranking ProblemExample = Document Retrieval Ranking Systems Documents query ranked list of documents * Learning to Rank for Information Retrieval * Ranking System Labels: 1) binary, 2) multiple-level, discrete, 3) pairwise preference, 4) Partial order or even total order of documents queries documents Training Data Test data Model Learning System min