摘要
相比规范新闻文本中命名实体识别(named entity recognition,NER),中文社交媒体中命名实体识别的性能偏低,这主要受限于文本的规范性和标注语料的规模。近年来中文社交媒体的命名实体识别研究主要针对标注语料规模小这一问题,倾向于使用外部知识或者借助联合训练来提升最终的识别性能,但对社交媒体文本不规范导致的对文本自身蕴含特征的挖掘不够这一问题的研究很少。该文着眼于文本自身,提出了一种结合双向长短时记忆和自注意力机制的命名实体识别方法。该方法通过在多个不同子空间捕获上下文相关信息来更好地理解和表示句子结构,充分挖掘文本自身蕴含的特征,并最终提升不规范文本的实体识别性能。在Weibo NER公开语料上进行了多组对比实验,实验结果验证了方法的有效性。结果表明:在不使用外部资源和联合训练的情况下,命名实体识别的F1值达到了58.76%。
Named entity recognition(NER)in Chinese social media is less effective than in standard news mainly due to the normalization and the size of the existing annotated corpus.In recent years,research on named entity recognition in Chinese social media has tended to use external knowledge and joint training to improve performance due to the small size of the annotated corpus.However,there are few studies on mining entity recognition characteristics in social media.This article focuses on named entity recognition in text articles using a neural network model that combines bi-directional long short-term memory with a self-attention mechanism.This model extracts context information from different dimensions to better understand and represent the sentence structure and improve the recognition performance.Tests on the Weibo NER released corpus show that this method is more effective than previous approaches and that this method has a 58.76% F1-score without using external knowledge or joint learning.
引文
[1] PENG N Y,DREDZE M.Named entity recognition for Chinese social media with jointly trained embeddings[C]//Proceedings of the 2015Conference on Empirical Methods in Natural Language Processing.Lisbon,Portugal:Association for Computational Linguistics,2015:548-554.
[2] PENG N Y, DREDZE M. Improving named entity recognition for Chinese social media with word segmentation representation learning[J].arXiv preprint arXiv:1603.00786,2016.
[3] HE H F,SUN X.F-score driven max margin neural network for named entity recognition in Chinese social media[J].arXiv preprint arXiv:1611.04234,2016.
[4] HE H F,SUN X.A unified model for cross-domain and semi-supervised named entity recognition in Chinese social media[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence. San Francisco, USA:AAAI,2017:3216-3222.
[5] HE J Z,WANG H F.Chinese named entity recognition and word segmentation based on character[C]//Proceedings of the Sixth SIGHAN Workshop on Chinese Language Processing.Hyderabad,India,2008.
[6] GOLDBERG Y, LEVY O. Word2vec explained:Deriving Mikolov et al.s negative-sampling word-embedding method[J].arXiv preprint arXiv:1402.3722,2014.
[7] HOCHREITER S,SCHMIDHUBER J.LSTM can solve hard long time lag problems[C]//Proceedings of the 9th International Conference on Neural Information Processing Systems.Denver,Colorado,USA:MIT Press,1996:473-479.
[8]冯辉.视觉注意力机制及其应用研究[D].北京:华北电力大学(北京),2011.FENG H.Research on visual attention mechanism and its application[D]. Beijing:North China Electric Power University(Beijing),2011.(in Chinese)
[9] VASWANI A,SHAZEER N,PARMAR N,et al.Attention is all you need[C]//Advances in Neural Information Processing Systems.Long Beach:NIPS,2017:6000-6010.
[10]YAO L,TORABI A,CHO K,et al.Video description generation incorporating spatio-temporal features and a soft-attention mechanism[J].arXiv preprint arXiv:1502.08029,2015.
[11]HUANG Z H,XU W,YU K.Bidirectional LSTM-CRF models for sequence tagging[J].arXiv preprint arXiv:1508.01991,2015.
[12]洪铭材,张阔,唐杰,等.基于条件随机场(CRFs)的中文词性标注方法[J].计算机科学,2006,33(10):148-151,155.HONG M C,ZHANG K,TANG J,et al.A Chinese part-of-speech tagging approach using conditional random fields[J].Computer Science,2006,33(10):148-151,155.(in Chinese)
[13]FORNEY G D.The viterbi algorithm[J].Proceedings of the IEEE,1973,61(3):268-278.
[14]BOTTOU L. Stochastic gradient descent tricks[M]//MONTAVON G, ORR G B, MLLER K R. Neural networks:Tricks of the trade. Berlin, Heidelberg,Germany:Springer,2012:421-436.
[15]陈治纲,何丕廉,孙越恒,等.基于向量空间模型的文本分类系统的研究与实现[J].中文信息学报,2005,19(1):37-41.CHEN Z G,HE P L,SUN Y H,et al.Research and implementation of text classification system based on VSP[J].Journal of Chinese Information Processing,2005,19(1):37-41.(in Chinese)