【摘要】 在微博情感倾向性分析中,一种典型分析方法是先对微博进行主客观分类,再对判定为主观的微博进行褒贬分类,但其问题在于主客观分类错误将直接传导到褒贬分类。针对这一问题,本文提出了一个主客观分类和褒贬分类融合的评估情感倾向性强度的模型。首先使用改进的逻辑回归模型构建主客观分类模型,并结合情感词典构建褒贬分类模型;然后,将二者融合,构建情感倾向性强度模型来选出具有较强情感的微博;最后应用褒贬分类模型判定情感倾向性。该方法在第六届中文倾向性分析评测(COAE2014)的微博观点句识别任务中获得了主要指标Micro_F1值和Macro_F1值的第二名。更多还原

【Abstract】 A typical practice in sentiment analysis consists of two steps:first classify the subjective sentences from the objective ones,and then distinguish the positives from the negatives among the subjective sentences.To alleviate the issue of error accumulation arising from such a pipeline approach,this paper investigates a unified model for microblog sentiment analysis.Firstly,a subjective-objective classification model is constructed by the improved Logistic Regression model.And a positive-negative classification model is proposed by using sentiment dictionary and the improved Logistic Regression.Secondly,an emotional intensity model,which is a linear combination of the two classification sub-models,is applied to select the microblogs with more strong sentiment.Lastly,the sentiment classification is judged by the positive-negative classification model.The final release of COAE 2014 indicates that the proposed method ranks as top 2 in micro_F1 and macro_F1 in the task.更多还原