A Neural Multi-digraph Model for Chinese NER with Gazetteers

Abstract

Gazetteers were shown to be useful resources for named entity recognition (NER) (Ratinov and Roth, 2009). Many existing approaches to incorporating gazetteers into machine learning based NER systems rely on manually defined selection strategies or handcrafted templates, which may not always lead to optimal effectiveness, especially when multiple gazetteers are involved. This is especially the case for the task of Chinese NER, where the words are not naturally tokenized, leading to additional ambiguities. To automatically learn how to incorporate multiple gazetteers into an NER system, we propose a novel approach based on graph neural networks with a multidigraph structure that captures the information that the gazetteers offer. Experiments on various datasets show that our model is effective in incorporating rich gazetteer information while resolving ambiguities, outperforming previous approaches.

A Mixed Semantic Features Model for Chinese NER with Characters and Words

“…Zhang et al [26] investigate a lattice network which explicitly leverages word and word sequence information, and achieve F1-score of 58.79%. Our proposed model has a significant improvement in the named entities, which improves 1.96% compared with Ding et al [5]. And overall performance is significantly better than other models.…”

“…[14] add gazetteer-enhanced sub-tagger on hybrid semi-Markov CRF architecture and observe some promising results. And [5] also propose a neural multi-digraph model with the information of gazetteers.…”

“…And Cao et al [2] also use the information of CWS for NER. Zhang et al [24] and Ding et al [5] add additional features, and the latter achieve 94.4% F1-score. Zhu et al [28] investigate a Convolution Attention Network to capture the information from adjacent characters and sentence contexts, which achieves F1-score of 92.97%.…”

Supervised machine learning for classification mining and ranking of illegal web contents = Aprendizaje automático supervisado para la clasificación, extracción de conocimiento y ordenación de contenidos web ilegales

“…During the last few years, researchers have improved the F1 score on this dataset using various techniques, ranging from using Bi-LSTM and CRF, a model similar to the design of Lample et al, and achieved 40.42% (Lin et al, 2017), incorporating the Transfer Learning (TL) approach that achieved an F1 score of 40.78%, to the model of Aguilar et al (2017) who boosted their model with an extra feature extracted from an external data resource, a gazetteer, and they scored an entity and surface F1 scores of 41.86% and 40.24%, respectively. Several researchers incorporated gazetteers to capture further features from the input text (Mishra and Diesner, 2016;Aguilar et al, 2017;Štravs and Zupančič, 2019;Dey and Prukayastha, 2013;Ding et al, 2019). Nevertheless, its usage is considered a limitation due to the difficulties of building and maintaining it up-to-date to cope with new terms and entities.…”

“…Varios investigadores han utilizado diccionarios geográficos para capturar características adicionales del texto de entrada (Mishra and Diesner, 2016;Aguilar et al, 2017;Štravs and Zupančič, 2019;Dey and Prukayastha, 2013;Ding et al, 2019). Sin embargo, su uso se podría considerar una limitación debido a las dificultades para construirlo y mantenerlo actualizado para hacer frente a nuevos términos y entidades.…”