Deep JEDi: Deep Joint Entity Disambiguation to Wikipedia for Russian

Autor:	Andrey Sysoev, Irina Nikishina
Rok vydání:	2019
Předmět:	Structure (mathematical logic) Sequence business.industry Computer science computer.software_genre Sequence labeling Meaning (philosophy of language) Entity linking Knowledge base Artificial intelligence Architecture Joint (audio engineering) business computer Natural language processing
Zdroj:	Lecture Notes in Computer Science ISBN: 9783030373337 AIST
Popis:	Over the past few years there has been a leap forward in both Entity Disambiguation and Entity Linking tasks. Meanwhile, Entity Disambiguation for Russian still lags behind advanced neural approaches developed for other languages. This paper introduces Deep JEDi—purely neural architecture, intended to identify the correct meaning for each mention in text. Combining sequence translation and sequence labeling approaches, our model achieves promising results on the Russian Wikipedia dataset. Significant improvement of its performance is attained by specific decoder that incorporates information about target mention position into attention mechanism. Additionally, we compare different approaches for learning distributed representations for tokens and entities and prove the importance of enriching joint embeddings with information about knowledge base structure.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::0fdaf825d0783ca606da75f095fa6291 https://doi.org/10.1007/978-3-030-37334-4_21 Zobrazit plný text záznamu