Poset representation and similarity comparisons os systems in IR

TL;DR

This paper introduces a poset-based approach to represent and compare complex IR system answers, incorporating clustering and ranking, and proposes new similarity measures for enhanced comparison accuracy.

Contribution

It presents a novel poset representation framework and a general method for constructing similarity measures that consider both clustering and ranking in IR answers.

Findings

Poset representation effectively models complex IR answers.

New similarity measures account for clustering and ranking.

Similarity indicator increases with common answer ranks.

Abstract

In this paper we are using the poset representation to describe the complex answers given by IR systems after a clustering and ranking processes. The answers considered may be given by cartographical representations or by thematic sub-lists of documents. The poset representation, with the graph theory and the relational representation opens many perspectives in the definition of new similarity measures capable of taking into account both the clustering and ranking processes. We present a general method for constructing new similarity measures and give several examples. These measures can be used for semi-ordered partitions; moreover, in the comparison of two sets of answers, the corresponding similarity indicator is an increasing function of the ranks of presentation of common answers.