A metric-based framework for automatic taxonomy induction

Abstract

This paper presents a novel metric-based framework for the task of automatic taxonomy induction. The framework incrementally clusters terms based on ontology metric, a score indicating semantic distance; and transforms the task into a multi-criteria optimization based on minimization of taxonomy structures and modeling of term abstractness. It combines the strengths of both lexico-syntactic patterns and clustering through incorporating heterogeneous features. The flexible design of the framework allows a further study on which features are the best for the task under various conditions. The experiments not only show that our system achieves higher F1-measure than other state-of-the-art systems, but also reveal the interaction between features and various types of relations, as well as the interaction between features and term abstractness.

Learning Pretopological Spaces for Lexical Taxonomy Acquisition

“…Note that this situation is reversed for the Plants and Food domains, which indicates that the subsumption relation can be described differently depending on the studied domain. This is an important issue when compared to [18,16], who suppose that the is-a relation can universally be learned from WordNet. Expectedly, N kN C shows poor performance due to its non-taxonomic nature.…”

“…In order to take advantage of multiple criteria, two important works have been proposed [16,18]. Both methodologies first learn an ontology metric, which models the is-a relation based on vectors of discriminant criteria (e.g.…”

“…This step is obtained by supervised learning over existing taxonomies. The logistic regression is used by [16] and [18] applies the ridge regression. Then, the ontology metric guides the incremental taxonomy acquisition process modeled as an optimization task: 1-objective for [16] and 2-objectives for [18].…”

“…The logistic regression is used by [16] and [18] applies the ridge regression. Then, the ontology metric guides the incremental taxonomy acquisition process modeled as an optimization task: 1-objective for [16] and 2-objectives for [18]. The main advantage of these approaches is to model the is-a relation between terms based on multiple criteria, thus greatly avoiding data sparseness and low coverage.…”

“…(1) We learn a propagation metric, which directly models the is-a relation into the taxonomy induction process in contrast to [18] and [16] who propose a two-steps process, (2) Linguistic patterns, which embody small (yet accessible) pieces of knowlegde of the domain constrain the semi-supervised learning process but are also used as relevant criteria, (3) We deal with both general and specialized domains where linguistic patterns fail to retrieve any relation, (4) Our framework is quasi-independent regarding to language as only few and simple linguistic patterns and raw texts are required.…”