基于多特征的微博情感分析研究

发布时间：2018-03-20 01:27

本文选题：微博　切入点：情感词　出处：《计算机工程》2017年12期 　论文类型：期刊论文

【摘要】：为提高微博情感分类识别的正确率,以网络微博数据作为研究对象,提出一种基于图的情感基准词选择方法。结合知网相似度知识,构建图模型,以图中节点中介性的值为依据,选择出高质量和高覆盖率的情感基准词。根据得到的基准词构建情感分析中所需的情感词典,并给出情感词极性。同时将情感词应用于挖掘短句情感特征,加入到传统支持向量机(SVM)模型中,对微博句子挖掘更多的语义信息从而获取更合理的语义合成函数,捕捉句子情感变化以更好地把握微博整句情感。采用具有特征约束特性的条件随机场(CRF)模型对短句进行分类。实验结果验证了CRF模型短句分类的有效性,与多种特征的SVM分类方法相比,在不同数据集上具有更好的分类效果。
[Abstract]:In order to improve the correct rate of Weibo's emotion classification and recognition, a graph-based affective benchmark word selection method is proposed, which is based on the online Weibo data, and the graph model is constructed by combining the knowledge of the knowledge of knowledge and network similarity. Based on the intermediate values of the nodes in the map, the emotional reference words with high quality and high coverage are selected. The emotion dictionary needed in the emotional analysis is constructed according to the obtained reference words. The polarity of affective words is also given. At the same time, affective words are applied to mining the affective features of short sentences and are added to the traditional support vector machine (SVM) model to mine more semantic information for Weibo sentences in order to obtain more reasonable semantic composition functions. In order to grasp Weibo's whole sentence emotion better, we use the conditional random field (CRF) model with characteristic constraints to classify short sentences. The experimental results show that the CRF model is effective in short sentence categorization. Compared with the SVM classification method with various features, it has better classification effect on different data sets.
【作者单位】：武汉大学计算机学院;武汉大学软件工程国家重点实验室;
【分类号】：TP391.1

【相似文献】