Fully-Unsupervised Embeddings-Based Hypernym Discovery

Abstract

The hypernymy relation is the one occurring between an instance term and its general term (e.g., “lion” and “animal”, “Italy” and “country”). This paper we addresses Hypernym Discovery, the NLP task that aims at finding valid hypernyms from words in a given text, proposing HyperRank, an unsupervised approach that therefore does not require manually-labeled training sets as most approaches in the literature. The proposed algorithm exploits the cosine distance of points in the vector space of word embeddings, as already proposed by previous state of the art approaches, but the ranking is then corrected by also weighting word frequencies and the absolute level of similarity, which is expected to be similar when measuring co-hyponyms and their common hypernym. This brings us two major advantages over other approaches—(1) we correct the inadequacy of semantic similarity which is known to cause a significant performance drop and (2) we take into accounts multiple words if provided, allowing to find common hypernyms for a set of co-hyponyms—a task ignored in other systems but very useful when coupled with set expansion (that finds co-hyponyms automatically). We then evaluate HyperRank against the SemEval 2018 Hypernym Discovery task and show that, regardless of the language or domain, our algorithm significantly outperforms all the existing unsupervised algorithms and some supervised ones as well. We also evaluate the algorithm on a new dataset to measure the improvements when finding hypernyms for sets of words instead of singletons.

Inferring the densest multi-profiled cross-community for a user

“…Zhang et al [21] proposed a method called TaxoGen to detect fine-grained topics using the combination of local-corpus embedding and spherical clustering. Atzori et al [22] proposed an unsupervised NLP-based method called HyperRank for detecting hypernyms from a corpus. This method employs the cosine distance between words in the vector space and does not require manually-labeled training sets.…”

RUG-1-Pegasussers at SemEval-2022 Task 3: Data Generation Methods to Improve Recognizing Appropriate Taxonomic Word Relations

“…The task of detecting taxonomic word relations has been tried via various approaches. From purely rule-based (Hearst, 1992), to using semantic treelike resources (Navigli et al, 2011), to adapting pre-trained language models (Atzori and Balloccu, 2020;Chen et al, 2021) or creating hybrid systems (Shwartz et al, 2016;Ravichander et al, 2020). Since this SemEval task is focused on neural (language) models, we opt to use BERT and focus mainly on different data generation techniques.…”

Knowledge Graphs Representation for Event-Related E-News Articles

“…Since knowledge representation in a KG generates a structured summary while highlighting the proximity of related concepts in a collection of text documents [2], KG can be used as an efficient application in representing knowledge in a set of e-news articles based around a particular news event. One of the most important linguistic relationships that can be found within a KG is the hierarchical relationship "hypernyms" [15,16]. For example, the "is-a" relation of conceptual ontologies relates subordinate words (e.g., "Sri Lanka") with its superordinate (e.g., "Country").…”

“…For example, the "is-a" relation of conceptual ontologies relates subordinate words (e.g., "Sri Lanka") with its superordinate (e.g., "Country"). Hierarchical relation lies at the origin of human reasoning and permits relating words representing specific instances to generic ones; moreover, it is one of extreme importance in any taxonomy [16].…”