基于组合神经网络的软件命名实体识别仿真  

Software Named Entity Recognition Simulation Based on Combined Neural Network

作　　者：卢青华[1] 袁丽娜[1] LU Qing-hua;YUAN Li-na(South China Institute of Software Engineering,Guangzhou University,Guangdong Guangzhou 510990,China)

机构地区：[1]广州大学华软软件学院,广东广州510990

出　　处：《计算机仿真》2023年第1期489-492,509,共5页Computer Simulation

基　　金：广东省“创新强校工程”科研项目(2018GXJK286)。

摘　　要：当前软件命名实体识别方法忽略了对命名实体标签的预测,存在精度等级(P@N)、F1值和KS值均偏低问题。提出基于组合神经网络的软件命名实体识别方法。将识别命名实体问题转化成“SBEIO”标签预测问题,在组合神经网络模型的基础上提取字、词特征,并将两者结合得到词向量特征,以此预测出最优标签序列。构建支持向量机分类器,将标签序列进行分类,根据超平面分割出分类结果,利用决策函数确定出最优分类命名实体,实现软件命名实体识别。实验结果表明,所提方法具有较高的精度等级(P@N)、F1值以及KS值。At present, some methods of named entity recognition often ignore the prediction of named entity tags, leading to low precision level(P@N),F1 value and KS value. Therefore, a method of named entity recognition of software based on a composite neural network was proposed. At first, the problem of recognizing the named entities was transformed into a problem of predicting "SBEIO" tags. Based on the composite neural network model, the character and word features were extracted and combined to get some word vector features, and thus to predict an optimal tag sequence. Then, a support vector machine classifier was constructed to classify label sequences. According to the hyperplane, the classification results were obtained. Finally, the decision function was used to determine an optimal named classification entity. Thus, the named entity recognition of software was achieved. Experimental results show that the proposed method has a high precision level(P@N),F1 value, and KS value.

关 键 词：词特征 字词向量 组合神经网络模型 支持向量机 命名实体识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]