【摘要】 文本情感分析是自然语言处理的重要部分,但现有的文本情感分析方法均有其不足.为了使各个方法进行互补,提出了一种融合改进Stacking与规则的文本情感分析方法 Stacking-I.该方法在Stacking集成算法的基础上进行改进,融合了两种主流的情感分析方法:文本规则方法和机器学习方法.在不同的3组网络评论文本上进行实验,证明该方法在网络评论文本情感分析实验中表现良好且有较高的准确率,其准确率高于传统机器学习方法、其它集成算法以及深度学习方法,最高可达91.700%,并且在不同数据量的基础上,通过大量实验和时间复杂度对比,得到了针对网络文本情感分析最佳的Stacking-I算法配置.更多还原

【Abstract】 Text emotion analysis is a vital part of natural language processing.However,the existing methods of text emotion analysis have their shortcomings.For the sake of combining the benefits of each model,a text emotion analysis method named Stacking-I based on the improved Stacking and text rules is developed.The algorithm is proposed on the basis of Stacking integration algorithm,combining two main emotion analysis methods:text rule method and machine learning method.Experiments were carried out on three different groups of network comment texts,which proved that this method has an excellent performance and a high accuracy rate in emotional analysis of network comment texts.Moreover,the highest accuracy rate reached 91.700% in the experiments,higher than traditional machine learning methods,other integrated algorithms and deep learning methods.By considering the time complexity and based on different datasets,the best algorithm of Stacking-I for emotion analysis of network text is configured after extensive experiments.更多还原