Learning Pretopological Spaces for Lexical Taxonomy Acquisition

Abstract

Abstract. In this paper, we propose a new methodology for semisupervised acquisition of lexical taxonomies. Our approach is based on the theory of pretopology that offers a powerful formalism to model semantic relations and transforms a list of terms into a structured term space by combining different discriminant criteria. In order to learn a parameterized pretopological space, we define the Learning Pretopological Spaces strategy based on genetic algorithms. In particular, rare but accurate pieces of knowledge are used to parameterize the different criteria defining the pretopological term space. Then, a structuring algorithm is used to transform the pretopological space into a lexical taxonomy. Results over three standard datasets evidence improved performances against state-of-the-art associative and pattern-based approaches.

Learning Pretopological Spaces to Model Complex Propagation Phenomena: A Multiple Instance Learning Approach Based on a Logical Modeling

“…These pretopological spaces have some good structuring properties, for example, the intersection of closed sets is a closed set ; Largeron and Bonnevay (2002) rely on these properties to define an algorithm that structures the elements of a V-type pretopological space (E, a) into a DAG. This algorithm is the cornerstone of the work of Cleuziou and Dias (2015) by considering a lexical taxonomy as a DAG structuring natural language terms. The V-type isotonic property will be also of crucial importance for the pseudo-closure learning framework we propose in the remaining.…”

“…Thus, research was carried out about a methodology aiming to build a pseudoclosure operator based on multiple neighborhoods and designed in such a way that useful neighborhoods are picked and combined together while erroneous ones are discarded. Cleuziou and Dias (2015) were the first tackling this problem by proposing to learn the combination of neighborhoods that defines the pseudo-closure operator (and hence its associated pretopological space). This section first recalls the (weighted) pseudo-closure definition proposed by Cleuziou and Dias (2015), then points the limitation of such a numerical and linear modeling and finally details a new proposal based on the logical formalism.…”

“…Our method actually learns the underlying pseudo-closure operator, which defines the pretopological space. Cleuziou and Dias (2015) were the first to tackle the problem of learning a pseudo-closure operator from observations. For a knowledge engineering purpose, they design a (semi)-supervised method to learn a pretopological space from which a lexical taxonomy is derived.…”

“…The present paper intends to both generalize and enhance the original approach from Cleuziou and Dias (2015). In this aim, three main contributions are introduced :…”

“…Finally, Section 6 is dedicated to quantitative and qualitative comparisons of our approach with the one presented in (Cleuziou and Dias, 2015) plus a greedy variation. We simulate a forest fire in a rectangular grid and we compare how each algorithm is capable of retrieving the underlying propagation model.…”