论文标题
关于排名公理对IR评估指标的影响
On the Effect of Ranking Axioms on IR Evaluation Metrics
论文作者
论文摘要
通过公理分析对IR评估指标的研究可以更好地了解其数值特性。一些作品已经建模了检索指标的有效性，其公理可以在文档排名集中捕获理想的属性。本文正式探讨了这些排名公理对某些IR评估指标数值的影响。它着重于具有多元文档相关性的一组排名的文档列表。该集合中的可能顺序来自检索指标上的三个普遍接受的排名公理。然后，它们通过其晶状体特性进行分类。当优先考虑相关文档时，确定了文档排名的一部分：可结合的元素，这些元素与向量空间中的基础概念具有一定的相似之处。可以计算从其值中的任何排名中的精度，回忆，RBP或DCG值中的精度，RBP或DCG值。但是，当考虑交换文件时，情况并非如此。
The study of IR evaluation metrics through axiomatic analysis enables a better understanding of their numerical properties. Some works have modelled the effectiveness of retrieval metrics with axioms that capture desirable properties on the set of rankings of documents. This paper formally explores the effect of these ranking axioms on the numerical values of some IR evaluation metrics. It focuses on the set of ranked lists of documents with multigrade relevance. The possible orderings in this set are derived from three commonly accepted ranking axioms on retrieval metrics; then, they are classified by their latticial properties. When relevant documents are prioritised, a subset of document rankings are identified: the join-irreducible elements, which have some resemblance to the concept of basis in vector space. It is possible to compute the precision, recall, RBP or DCG values of any ranking from their values in the join-irreducible elements. However this is not the case when the swapping of documents is considered.
