在线学习情感分类模型研究

Research on Sentiment Classification Model of Online Learning

本文结合Adadelta算法学习率自适应调整和Adam算法避免了训练后期频繁抖动的特点,提出了Adamdelta算法,解决了传统FTRL-Proximal在线学习算法学习率随着训练次数增加逐渐消失的问题。使用一阶和二阶矩估计进行偏差修正调整FTRL-Proximal算法学习率,再使用梯度下降求解模型权重参数,进而得到LR模型,并以此模型完成在线学习情感分类工作。为了验证改进算法的优越性,利用IMDB电影评论文本做实验数据,与5种模型进行对比分析。实验表明,改进算法具有更好的分类效果,有效的提高了分类器的准确率和召回率。

In this paper,Adamdelta algorithm is presented combining the characteristics that adaptive adjustment of learning rate for Adadelta algorithm and avoid frequent jitter in the later stage of training for Adam algorithm.The Adamdelta algorithm is proposed to solve the problem that learning rate of traditional FTRL-Proximal online learning algorithm will disappear with the increase of training times.First-order and second-order moment estimation are used to adjust learning rate of the FTRL-Proximal algorithm by deviation correction,then gradient descent is used to solve the model weight parameters.Moreover,the LR model is obtained which used to classification of online learning emotional.In order to verify the superiority of the improved algorithm,the five models are compared and analyzed by using the experimental data of the IMDB movie review text.Experiments show that the improved algorithm has better classification effect and effectively improves accuracy and recall rate of the classifier.