基因遗传算法在文本情感分类中的应用

Application of genetic algorithm in text sentiment classification

摘要:

本文以微博文本为主要实验对象, 提出适合卷积神经网络进行自我优化的编码方式，分别将每一层看做是一个染色体，将每一层中的参数看做是一个基因片段，采用混合双重非数值编码的方式编码每个CNN框架，设计出适合于CNN网络的选择、交叉和变异的算法，并且把基因遗传算法(GA)和与卷积神经网络相结合,提出了基于情感分析算法的遗传算法(GA-CNN)。通过对传统算法与GA-CNN的实验与对比分析，良好地展示了自我优化性。

Abstract:

In this paper, we use Sina Weibo text as the main experimental dataset, and propose a coding method suitable for selfoptimization of convolutional neural networks. Our coding method encodes each CNN framework using a hybrid double nonnumeric encoding by treating each layer as a chromosome and the parameters in each layer as a gene segment respectively, and selection, crossover and mutation algorithms are devised for CNN networks. We also propose a genetic algorithm based on sentiment analysis algorithm (GACNN) which combines genetic algorithm (GA) with convolutional neural network. The experiment and comparative analysis of GACNN and traditional algorithms demonstrates the selfoptimization of our method.