Named Entity Recognition with Context-Aware Dictionary Knowledge

Named Entity Recognition with Context-Aware Dictionary Knowledge Chuhan Wu†, Fangzhao Wu‡, Tao Qi†, Yongfeng Huang† †Department of Electronic Engineering & BNRist, Tsinghua University, Beijing 100084, China ‡Microsoft Research Asia, Beijing 100080, China {wuchuhan15, wufangzhao, taoqi.qt}@gmail.com yfhuang@tsinghua.edu.cn Abstract Named entity recognition (NER) is an important task in the natural language processing field. Existing NER methods heavily rely on labeled data for model training, and their performance on rare entities is usually unsatisfactory. Entity dictionaries can cover many entities including both popular ones and rare ones, and are useful for NER. However, many entity names are context- dependent and it is not optimal to directly apply dictionaries without considering the context. In this paper, we propose a neural NER approach which can exploit dictionary knowledge with contextual information. We propose to learn context-aware dictionary knowledge by modeling the interactions between the entities in dictionaries and their contexts via context-dictionary attention. In addition, we propose an auxiliary term classification task to predict the types of the matched entity names, and jointly train it with the NER model to fuse both contexts and dictionary knowledge into NER. Extensive experiments on the CoNLL-2003 benchmark dataset validate the effectiveness of our approach in exploiting entity dictionaries to improve the performance of various NER models. 1 Introduction Named entity recognition (NER) aims to extract entity names from texts and classify them into several pre-defined categories, such as person, location and organization (Levow, 2006). It is an important task in natural language processing, and a prerequisite for many downstream applications such as entity linking (Derczynski et al., 2015) and relation extraction (Lin et al., 2016; Luo et al., 2018; Zeng et al., 2018). Thus, NER is a hot research topic. In this paper, we focus on the English NER task. Many methods have been proposed for English NER, and most of them model this task as a word-level sequence labeling problem (Chiu and Nichols, 2016). For example, Ma and Hovy (2016) proposed a CNN-LSTM-CRF model for English NER. They used CNN to learn word representations from charac- ters, LSTM to model the contexts of words, and CRF to decode labels. These existing NER methods usually rely on massive labeled data for model training, which is costly and time-consuming to annotate. When training data is scarce, their performance usually significantly declines (Peng et al., 2019). In addition, their performance on recognizing entities that rarely or do not appear in training data is usually unsatisfactory (Wang et al., 2019). Fortunately, many large-scale entity dictionaries such as Wikipedia (Higashinaka et al., 2012) and Geonames1 are off-the-shelf, and they can be easily derived from knowledge bases and webpages (Nee- lakantan and Collins, 2014). These entity dictionaries contain both popular and rare entity names, and can provide important information for NER models to identify these entity names. There are a few researches on incorporating entity dictionary into NER (Liu et al., 2019; Magnolini et al., 2019) and most of them are based on dictionary matching features. For example, Wang et al. (2019) proposed …

Content truncated to 500 words