一种基于动态协同神经网络的文本作者身份分类研究

Dynamic Synergetic Neural Network Algorithm for Authorship Classification of Texts

为了提高文本作者身份分类的准确度,提出一种动态协同神经网络算法。该算法利用了协同神经网络训练速度快、抗造声强等特点,并采取了注意参数动态调整的策略。通过原型模式向量与实验模式向量间的相似性动态地选取合适的注意参数,在演化过程中对误识别的模式进行自适应纠正。与平衡注意参数条件下的识别效果进行对比校验,结果表明,该算法在很大程度上提高了网络的自学习能力,从而改善了作者身份分类的精度和鲁棒性。

To improve the accuracy of authorship classification of texts,a dynamic synergetic neural network algorithm was proposed in this study.Specifically,this algorithm makes use of the characteristics of fast training and strong noiseimmunity,and dynamically adjusts the attention parameters.Thus,the initial mis-identified patterns are adaptively corrected by measuring similarity between the prototype pattern and the testing pattern in evolution process.Compared with classification result under balanced attention parameter,the experimental result demonstrates that self-learning ability of the network is significantly improved in the dynamic synergetic neural network algorithm,thus the classification performance and robustness are improved.