Modeling Score Distributions

Modeling Score Distributions

Anca Doloc-Mihu (University of Louisiana at Lafayette, USA)
Copyright: © 2009 |Pages: 7
DOI: 10.4018/978-1-60566-010-3.ch206
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

The goal of a web-based retrieval system is to find data items that meet a user’s request as fast and accurately as possible. Such a search engine finds items relevant to the user’s query by scoring and ranking each item in the database. Swets (1963) proposed to model the distributions of these scores to find an optimal threshold for separating relevant from non-relevant items. Since then, researchers suggested several different score distribution models, which offer elegant solutions to improve the effectiveness and efficiency of different components of search systems. Recent studies show that the method of modeling score distribution is beneficial to various applications, such as outlier detection algorithms (Gao & Tan, 2006), search engines (Manmatha, Feng, & Rath, 2001), information filtering (Zhang & Callan, 2001), distributed information retrieval (Baumgarten, 1999), video retrieval (Wilkins, Ferguson, & Smeaton, 2006), kernel type selection for image retrieval (Doloc-Mihu & Raghavan, 2006), and biometry (Ulery, Fellner, Hallinan, Hicklin, & Watson, 2006). The advantage of the score distribution method is that it uses the statistical properties of the scores, and not their values, and therefore, the obtained estimation may generalize better to not seen items than an estimation obtained by using the score values (Arampatzis, Beney, Koster, & van der Weide, 2000). In this chapter, we present the score distribution modeling approach, and then, we briefly survey theoretical and empirical studies on the distribution models, followed by several of its applications.
Chapter Preview
Top

Background

The primary goal of information retrieval is to retrieve all the documents which are relevant to a user query, while retrieving as few non-relevant documents as possible (Baeza-Yates & Ribeiro-Neto, 1999). This is achieved by ranking the list of documents according to their relevance to the user’s query. Since relevance is a subjective attribute, depending on the user’s perception of the closeness between the user submitted query and the real query from her or his mind, building a better way to retrieve data is a challenge that needs to be addressed in a retrieval system.

In other words, a retrieval system aims at building the request (query) that best represents the user’s information need. This optimal request is defined by using an explicit data-request matching (Rocchio, 1971) that should produce a ranking in which all relevant data are ranked higher than the non-relevant data. For the matching process, a retrieval system uses a retrieval function, which associates each data-query pair with a real number or score (the retrieval status value). Then, the retrieval system uses these scores to rank the list of data.

However, researchers (Swets, 1963; Arampatzis, Beney, Koster, & van der Weide, 2000; Manmatha, Feng, & Rath, 2001) raised the question of whether or not the statistical properties of these scores, displayed by the shape of their distribution, for a given query, can be used to model the data space or the retrieval process. As a result, they proposed and empirically investigated several models of the score distributions as solutions to improve the effectiveness and efficiency of the retrieval systems. The next section introduces the score distribution method.

Complete Chapter List

Search this Book:
Reset