论文部分内容阅读
微博情感倾向性分析通常指对中文微博中每个句子褒义、贬义或者中性的情感进行自动分类。针对微博碎片化和情感类别失衡的特点,在半监督学习reserved self-training方法的框架基础上提取了适用于微博情感分类的文本特征,并提出了针对情感倾向性分析通过训练度阈值设定的方法来优化reserved self-training迭代终止的条件,在保留reserved self-training能有效处理微博语料中语料情感不平衡问题的优点基础上,防止了训练过度情况的发生。COAE 2014微博情感倾向性评测结果证明了该方法的有效性。
Weibo affective tendencies analysis usually refers to the Chinese microblogging each sentence commendatory, derogatory or neutral emotion automatically classified. According to the characteristics of microfibre fragmentation and the imbalance of affective categories, text features suitable for the emotional classification of Weibo are extracted based on the framework of the semi-supervised reserved self-training method. Aiming at the analysis of affective tendencies, The method to optimize the condition of the end of the reserved self-training iteration is based on the precondition of preserving the reserved self-training that can effectively deal with the imbalance of the corpus emotion in the Weibo corpus, and prevent the over-training situation from occurring. The results of the COAE 2014 Weibo Emotional Evaluation proved the effectiveness of this method.