Coupled-Tensor Generated Word Embeddings and Their Composition

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F23%3A00367813" target="_blank" >RIV/68407700:21230/23:00367813 - isvavai.cz</a>

Výsledek na webu

<a href="https://doi.org/10.1007/978-3-031-37717-4_49" target="_blank" >https://doi.org/10.1007/978-3-031-37717-4_49</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-37717-4_49" target="_blank" >10.1007/978-3-031-37717-4_49</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Coupled-Tensor Generated Word Embeddings and Their Composition

Popis výsledku v původním jazyce

Contemporary methods of computing vector-space embeddings of words are able to accurately capture both their semantic and syntactic properties. Methods for computing n-gram embeddings do, however, come with downsides. They either require high resources during training or estimation, or come with other disadvantages, such as loss of information about individual positions of words in phrases. We propose two novel approaches to training word vectors enabling a composition of word embeddings into n-gram embeddings. Both methods are based on coupled CP decomposition of tensors that are generated by a sequence of time-shifted word embeddings. We compare our methods with SGNS and show that they provide superior performance on word-analogy tasks.

Název v anglickém jazyce

Coupled-Tensor Generated Word Embeddings and Their Composition

Popis výsledku anglicky

Contemporary methods of computing vector-space embeddings of words are able to accurately capture both their semantic and syntactic properties. Methods for computing n-gram embeddings do, however, come with downsides. They either require high resources during training or estimation, or come with other disadvantages, such as loss of information about individual positions of words in phrases. We propose two novel approaches to training word vectors enabling a composition of word embeddings into n-gram embeddings. Both methods are based on coupled CP decomposition of tensors that are generated by a sequence of time-shifted word embeddings. We compare our methods with SGNS and show that they provide superior performance on word-analogy tasks.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Lecture Notes in Networks and Systems

ISBN

978-3-031-37716-7

ISSN

2367-3370

e-ISSN

—

Počet stran výsledku

15

Strana od-do

753-767

Název nakladatele

Springer International Publishing AG

Místo vydání

Cham

Místo konání akce

Londýn

Datum konání akce

22. 6. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—