Social Media Named Entity Recognition Based On Graph Attention Network

Abstract

Named entity recognition is a basic task of natural language processing. With the development of the Internet, social media has become the main way of information transmission and sharing with friends, but social media texts are often short and informal, resulting in low entity recognition accuracy. In this paper, we use the Graph Attention Network (GAT) for the first time in the task of Chinese social media named entity recognition to generate the representation of the nodes in the sentence selection parsing tree, and capture the information of the sentence itself through the self-attention layer. In order to reduce the impact of Chinese word segmentation errors, we use a combination of character vectors and word vectors, and incorporate part of speech information for training. We integrate the above two points into the classic named entity recognition model BiLSTM-CRF model to form a new model to conduct research on named entity recognition of Chinese social media text. Experimental results show that our model achieves competitive results.