采用改进粒子群优化的SVM方法实现中文文本情感分类

近年来,随着网络用户量的不断增加,用户评论数量也呈爆炸式增长,伴随而来的是大量可用于参考和深度挖掘的信息,文本情感分类应运而生。分类模型的预测精度和执行速度是衡量模型优劣的关键。使用传统的SVM进行文本情感分类,算法简单,易于实现,但其模型参数决定了分类准确率。针对这种情况,文中将改进粒子群优化算法与SVM分类方法相结合,采用了改进粒子群算法优化的SVM方法对影视剧评论的情感进行了研究分析。首先,通过网络爬虫获取豆瓣电影评论数据,将数据预处理后利用加权word2vec向量化文本信息,将其作为支持向量机可识别的输入;然后,使用自适应惯性递减策略并引入交叉算子来改进粒子群算法,并对SVM模型的损失函数、惩罚参数及核函数的参数进行优化;最后,实现文本的情感分类。在同一数据集上的实验结果表明,所提方法有效规避了传统的情感词典方法受词语顺序和不同语境影响的缺陷及使用卷积出现梯度消失或弥散的问题,同时也克服了粒子群算法易陷入局部最优的不足。相较于其他方法,所提分类模型的执行速度更快,有效地提高了分类准确率。

Using SVM Method Optimized by Improved Particle Swarm Optimization to Analyze Emotion of Chinese Text

In recent years,with the increasing number of network users,the number of user comments has also increased explosively,accompanied by a large number of information that can be used for reference and deep excavation.Text sentiment classification arises at this historic moment,the prediction accuracy and the execution speed of classification model are the keys to mea-sure the quality of the model.Traditional algorithm by using SVM for text sentiment classification is simple and easy to implement,and its model parameters determine the classification accuracy.In this case,this paper combined the improved particle swarm optimization algorithm with the SVM classification method,used the SVM method optimized by improved particle swarm optimization to analyze the emotion of the movie and TV drama review.Firstly,Douban movie review data are obtained by internet crawler.Then the text information is vectorized by weighted word2vec after pre-processing,which becomes the recognizable input of support vector machine.Adaptive inertia decreasing strategy and crossover operator are used to improve particle swarm optimization algorithm.The loss function,penalty parameter and kernel parameter of SVM model are optimized by improved PSO.Finally,the text is classified by this model.Experimental results on the same data show that this method effectively avoids the shortcomings of traditional affective dictionary method affected by word order and different contexts,and solves the problem of gradient disappearance or dispersion caused by convolution.It also overcomes the possibility that PSO itself is easily trapped in local optimum.Compared with other methods,the proposed classification model performs faster and improves classification accuracy effectively.