Dhps: Dependency hearst's patterns for hypernym relation extraction
A Issa Alaa Aldine, M Harzallah, G Berio… - … : 10th International Joint …, 2020 - Springer
Knowledge Discovery, Knowledge Engineering and Knowledge Management: 10th …, 2020•Springer
Hearst's patterns are lexico-syntactic patterns that have been extensively used to extract
hypernym relations from texts. They are defined as regular expressions based on lexical and
syntactical information of each word. Here, we propose a new formulation of Hearst's
patterns using dependency parser, called Dependency Hearst's Patterns (DHPs). They are
defined as dependency patterns based on dependency relations between words. This
formulation allows us to define more generic Hearst's patterns that match better complex or …
hypernym relations from texts. They are defined as regular expressions based on lexical and
syntactical information of each word. Here, we propose a new formulation of Hearst's
patterns using dependency parser, called Dependency Hearst's Patterns (DHPs). They are
defined as dependency patterns based on dependency relations between words. This
formulation allows us to define more generic Hearst's patterns that match better complex or …
Abstract
Hearst’s patterns are lexico-syntactic patterns that have been extensively used to extract hypernym relations from texts. They are defined as regular expressions based on lexical and syntactical information of each word. Here, we propose a new formulation of Hearst’s patterns using dependency parser, called Dependency Hearst’s Patterns (DHPs). They are defined as dependency patterns based on dependency relations between words. This formulation allows us to define more generic Hearst’s patterns that match better complex or ambiguous sentences. To evaluate our proposal, we have compared the performance of Dependency Hearst’s patterns to lexico-syntactic patterns: Hearst’s patterns and an extended set of Hearst’s patterns applied on two corpora: Music and English. Dependency Hearst’s patterns yield to a considerable improve in term of recall and a slight decrease in term of precision.
Springer
Showing the best result for this search. See all results