基于差分进化的中文情感分类集成算法研究

Research on Ensemble Algorithm for Chinese Emotion Classification Based on Differential Evolution

情感分类是一种从文本中提取情感倾向的文本分类任务。集成学习通过结合几个分类器,在情感分类任务上能够获得比个体分类器更好的分类效果。但是,由于个体分类器在数据集上的表现不同,个体分类器在集成方法中的权重难以确定。针对集成学习中个体分类器的权重优化问题,提出一种基于差分进化优化个体分类器权重的集成分类方法,并将其应用于中文情感分类。以分类准确率为适应度值,通过差分进化算法优化5种个体分类器的权重组合,在3个领域的评论语料集上进行实验。实验结果表明,与一般的集成方法相比,该方法在中文情感分类上有更好的分类效果。

Sentiment classification is a text classification task that extracts emotional tendencies from text. Ensemble learning combines several classifiers to achieve a better classification effect than component learners on the task of emotional classification. However,due to the different performance of component learners on data sets,the weights of component learners in the ensemble method were not well-distributed. Aiming at the weight optimization problem of component learners in ensemble learning,a classification method based on differential evolution to optimize the weights of component learners was proposed and applied to Chinese text sentiment classification. Using the classification accuracy as the fitness value,the weight combination of five component learners was optimized by differential evolution algorithm,and experiments were performed on the review corpus of the three fields. The experimental results showed that compared with the general ensemble method,the proposed classification method has better classification performance in sentiment classification.