TAXI at SemEval-2016 Task 13: a Taxonomy Induction Method based on Lexico-Syntactic Patterns, Substrings and Focused Crawling

Abstract

We present a system for taxonomy construction that reached the first place in all subtasks of the SemEval 2016 challenge on Taxonomy Extraction Evaluation. Our simple yet effective approach harvests hypernyms with substring inclusion and Hearst-style lexicosyntactic patterns from domain-specific texts obtained via language model based focused crawling. Extracted taxonomies are evaluated on English, Dutch, French and Italian for three domains each (Food, Environment and Science). Evaluations against a gold standard and by human judgment show that our method outperforms more complex and knowledge-rich approaches on most domains and languages. Furthermore, to adapt the method to a new domain or language, only a small amount of manual labour is needed.

SemEval-2016 Task 13: Taxonomy Extraction Evaluation (TExEval-2)

“…TAXI The methods for hypernym identification used in the TAXonomy Induction system (TAXI) rely on two sources of evidence: substring matching Table 4: Manual evaluation of 100 (at most) randomly selected novel relations based on precision for English and Hearst-like patterns (Panchenko et al, 2016). The Hearst patterns for all languages are extracted from Wikipedia and from focused crawls with seed pages that are Wikipedia pages.…”

A Short Survey on Taxonomy Learning from Text Corpora: Issues, Resources and Recent Advances

“…In both task, two pattern-based methods (i.e., IN-RIASAC (Grefenstette, 2015) in TExEval and TAXI (Panchenko et al, 2016) in TExEval-2) consistently outperform others. INRIASAC uses frequency-based co-occurrence statistics, and substring inclusion heuristics to extract a set of hypernyms for hyponyms.…”

Taxonomy Induction Using Hypernym Subsequences

“…• Normalized Frequency Diff (n d ): Similar to [28], this feature is an asymmetric hypernymy score based on frequency counts. We compute n d (x i , x j ) by first normalizing the frequency counts obtained (i.e., the counts in E k (x i )) for term x i as follows:…”

“…Past approaches to taxonomy induction from scratch either assume the availability of a clean input vocabulary [28] or employ a time-consuming manual cleaning step over a noisy input vocabulary [38]. For example, Figure 1 shows the pipeline of a typical taxonomy induction approach from a domain corpus [38].…”