基于特征扩展与深度学习的短文本情感判定方法

发布时间：2018-05-24 20:22

本文选题：情感挖掘 + 短文本　；参考：《计算机科学》2017年10期

【摘要】：针对中文短文本信息量少、特征稀疏等特点,面向微博短文本进行情感分类研究,为了更好地提取短文本情感特征,从评论转发等上下文内容中挖掘具有语义递进关系的语料对原文本进行扩展,并抽取具有潜在感情色彩的特征词,采用Word2vec计算词语相似度以进行候选特征词扩展,最后引入深度信念网络(Deep Belief Network,DBN)对候选特征词进行深度自适应学习。在COAE(Chinese Opinion Analysis Evaluation)2015任务评测数据集上的实验表明,该方法能够有效地缓解短文本特征稀疏问题,并且能够较为准确地挖掘情感特征,提高情感分类的准确率。
[Abstract]:In view of the characteristics of Chinese short text, such as less information and sparse features, the emotion classification research is carried out for Weibo short text, in order to extract the emotion feature of short text better, The original text is extended by mining the corpus with semantic progressive relation from the context such as comment and forwarding, and the feature words with potential emotional color are extracted. Word2vec is used to calculate the similarity of words in order to extend the candidate feature. Finally, a deep belief network (Deep Belief Network DBN) is introduced to study candidate features in depth adaptive learning. Experiments on the COAE(Chinese Opinion Analysis Evaluation)2015 task evaluation data set show that the proposed method can effectively alleviate the sparse problem of short text features and can mine emotional features accurately and improve the accuracy of emotion classification.
【作者单位】：北京工业大学计算机学院;
【基金】：国家科技支撑计划子课题(2013BAH21B02-01) 北京市自然科学基金资助项目(4153058) 上海市智能信息处理重点实验室开放基金(IIPL-2014-004)资助
【分类号】：TP391.1

【相似文献】