融合神经网络与全局推理的实体共指消解算法  

Algorithm for Entity Coreference Resolution with Neural Network and Global Reasoning

作　　者：周宁 靳高雅 石雯茜 Zhou Ning;Jin Gaoya;Shi Wenqian(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《数据分析与知识发现》2022年第8期75-83,共9页Data Analysis and Knowledge Discovery

基　　金：国家自然科学基金项目(项目编号:61650207);兰州交通大学天佑创新团队(项目编号:TY202003)的研究成果之一。

摘　　要：【目的】提出融合神经网络与全局推理的实体共指消解模型,解决文本内实体信息复杂,指代信息具有不明确性且分布稀疏的问题,探索更有效的共指消解算法。【方法】首先,利用神经网络模型抽取出文档中的实体和其前指词;其次,结合句子的上下文信息进行全局推理,将此推理结果加入神经网络模型中,从而提高实体共指消解的精确度。【结果】在OntoNotes 5.0数据集上进行实体共指消解实验,结果验证了所提算法的有效性。融合神经网络与全局推理的实体共指消解算法能有效地提高共指消解性能和更好地理解文本语义信息,最终在CoNLL评测标准下F1值达到74.76%。【局限】需加入更精确的知识推理。【结论】与近几年其他的共指消解模型对比实验结果证明了所提算法的实用性与有效性。[Objective]This paper proposes a model for entity coreference resolution,which integrates neural network and global reasoning.It tries to address the issues of complex entity information in the text as well as the ambiguity and sparse distribution of referential information.[Methods]First,we used the neural network model to extract the entities and their antecedents from the documents.Then,we combined the context information of the sentence to perform global reasoning.Finally,we added the reasoning results to the neural network model to improve the accuracy of entity coreference resolution.[Results]We examined our new model on the OntoNotes 5.0 dataset.The new model’s F1 score reached 74.76%under the CoNLL evaluation standard.[Limitations]More precise knowledge reasoning needs to be added.[Conclusions]Compared with the existing models,the proposed algorithm improves the coreference resolution performance and better understand text semantic information.

关 键 词：神经网络 共指消解 实体消歧 全局推理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]