Query modeling for entity search based on terms, categories, and examples

Abstract

Users often search for entities instead of documents, and in this setting, are willing to provide extra input, in addition to a series of query terms, such as category information and example entities. We propose a general probabilistic framework for entity search to evaluate and provide insights in the many ways of using these types of input for query modeling. We focus on the use of category information and show the advantage of a category-based representation over a term-based representation, and also demonstrate the effectiveness of category-based expansion using example entities. Our best performing model shows very competitive performance on the INEX-XER entity ranking and list completion tasks.

Biography-Dependent Collaborative Entity Archiving for Slot Filling

“…Since INEX was launched in 2002, which is an entity ranking task specific to structured data and multimedia (Demartini et al, 2010), structured features have been widely used in entity search, such as the most recent studies on Wikipedia links and categories (Vercoustre et al, 2008;Zhu et al, 2008;Jiang et al, 2009;Weerkamp et al, 2009;Kaptein and Kamps, 2009;Balog et al, 2011) and web link structure (Balog et al, 2008b;You et al, 2011;Blanco et al, 2011;Neumayer et al, 2012;Bron et al, 2013).…”

Semistructured Data Search

“…Therefore, simply filtering on the target categories is insufficient, more robust techniques are needed. A standard way of using category information is to include a separate category similarity component in the overall ranking formula [6,16,22,35]. A principled way realizing this idea is proposed in [6], where both documents and queries have a dual representation, one based on terms and one based categories.…”

“…A standard way of using category information is to include a separate category similarity component in the overall ranking formula [6,16,22,35]. A principled way realizing this idea is proposed in [6], where both documents and queries have a dual representation, one based on terms and one based categories. Formally, each document is represented as a pair, d = (θ Figure 12, where the left and right sides symbolize the query and the document, respectively, where each have a term-based (top) and a category-based (bottom) representation, modeled as multinomial probability distributions.…”

“…This does not make a difference when a single representation is used, but becomes an issue when term-based and category-based components need to be combined. Simply put, this transformation ensures that the probabilities are computationally tractable and "compatible;" the interested reader is referred to [6] for further details. What remains to be defined, then, is the estimation of the ingredients for category-based similarity.…”

“…Techniques introduced in Section 5.1 (blind relevance feedback and sampling from example documents) can be adopted in a straightforward way to categories. In fact, it has been shown that category-based feedback can be more beneficial than term-based feedback [6]. There exist further possibilities specific to categories.…”