Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

时间:2023-11-22 12:57:02

读了一篇paper,MSRA的Wei Wu的一篇《Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata》。是关于Ranking Relevence方面的文章。下面简单讲下我对这篇文章的理解,对这方面感兴趣的小伙伴们可以交流一下。

1. Abstract

这篇文章的重点在于使用query-doc的点击二部图,结合query/doc的meta数据(组织成multiple types of features),来学习出query-doc(顺带介绍了query-query,doc-doc)的similarity

为了计算上述的similarity,作者采用了两个不同的linear mappings,用来把query从query feature space,把doc从doc feature space映射到相同的latent space上,然后便可通过计算这个latent space上两者的vector的dot product来获得两者的similarity。于是,便把对similarity的learning形式化为对mapping的learning,而这个mapping的learning的目标是为了maximize从enriched click-through bipartite gragh上观察到的query-doc的similarity(可以通过query-doc pair的点击数来衡量)。另外,这个linear mapping是针对一种类型的features,获得一种类型features的similarity function,如果有multiple types of features的话,则最终的similarity function是每个type的similarity function的线性组合。

learning过程用到的算法包括Singular Value Decomposition(SVD)和Multi-view Partial Least Squares(M-PLS)。

2. Introduction

作者提到了先前的关于计算query-doc similarity的几种方法。

1)feature based methods:Vector Space Model(VSM),BM25,Language Models for Information Retrieval(LMIR)等。

2)gragh based methods:mining query-doc similarity from a click-through bipartite gragh等。

而这篇文章是将两者结合起来:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

3. Problem Formulation

将每种type的features的query或者document用一个向量的形式来表示,,则linear mapping可以看做是维度为Learning Query and Document Similarities from Click-through Bipartite Graph with MetadataLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata的两种形式的矩阵(Learning Query and Document Similarities from Click-through Bipartite Graph with MetadataLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata ),通过这两种变换矩阵,query或者doc在原始空间上的向量被变换成latent space上的维度为Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata的向量Learning Query and Document Similarities from Click-through Bipartite Graph with MetadataLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata。于是,对于这种type的faetures,simialrity function表示为Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata。我们可以将点击二部图中query-doc的点击数看作是query-doc similarity的大小,而通过maximize观察到的query-doc的similarity来学习linear mappingLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata和线性加权的权重Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

最终的learning problem可以表示为:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

这时候有个问题,就是需要最大化的公式的值是可以无限大的,因为没有系数的限制,下面会介绍如何在系数上加上constraints。

4. Multi-view Partial Least Squares

 4.1 Constrained Optimization Problem

1)对feature vectors进行归一化:Learning Query and Document Similarities from Click-through Bipartite Graph with MetadataLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata

2)对mapping matricesLearning Query and Document Similarities from Click-through Bipartite Graph with Metadata进行正交化限制。

3)对线性加权权重Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata进行L2 正则化限制。

于是,learning method重新形式化为:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

 4.2 Globally Optimal Solution

为了获得全局最优解,两步走。第一步,对每种type的features,通过SVD求解得到optimal linear mapping;第二步,求解optimal combination weights。

上述的公式(2)可以重写为:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

optimization problem为:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

通过SVD求得global optimal solution。

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

于是,公式(2)可以写成:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

而combination weights求解为:

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

4.3 Learning Algorithm

1)for each type of feature,solves SVD of Mi to learn the linear mapping。

2)calculates the combination weights using (5)。

Learning Query and Document Similarities from Click-through Bipartite Graph with Metadata

版权声明:

本文由笨兔勿应所有,发布于http://www.cnblogs.com/bentu*ng。如果转载,请注明出处,在未经作者同意下将本文用于商业用途,将追究其法律责任。