基于词语相关度的微博新情感词自动识别

发布时间：2019-06-06 15:11

【摘要】：针对微博中新情感词的识别问题,提出了一种基于词语相关度的微博新情感词自动识别方法。首先,对于分词软件把一个新词错分成几个词的问题,利用组合思想将相邻词进行合并作为新词的候选词;其次,为了充分利用词语上下文的语义信息,采用神经网络训练语料获得候选新词的空间表示向量;最后,利用已有的情感词典作为指导,融合基于词表集合的关联度排序和最大关联度排序算法,在候选词上筛选,获得最终的情感新词。在COAE2014(第六届中文倾向性分析评测)任务3语料上,提出的融合算法与点互信息(PMI)、增强互信息(EMI)、多词表达距离(MED)、新词语概率(NWP)以及基于词向量的新词识别方法相比,准确率至少提高了22%,说明该方法自动识别微博新情感词效果优于其他五种方法。
[Abstract]:In order to solve the problem of recognition of new emotional words in Weibo, an automatic recognition method of Weibo new emotional words based on word correlation degree is proposed. First of all, for the problem that a new word is misdivided into several words by word segmentation software, the adjacent words are merged as candidates for new words by using the combination idea. Secondly, in order to make full use of the semantic information of word context, neural network is used to train the corpus to obtain the spatial representation vector of candidate new words. Finally, using the existing emotion dictionary as the guide, the correlation degree sorting and the maximum correlation degree sorting algorithm based on thesaurus set are combined to screen the candidate words and obtain the final emotional new words. Based on the task 3 corpus of COAE2014 (sixth Chinese tendency Analysis and Evaluation), the proposed fusion algorithm and point mutual information (PMI), enhance the distance (MED), of (EMI), multi-word expression with mutual information. Compared with the new word probability (NWP) and the new word recognition method based on word vector, the accuracy of the new word recognition method is improved by at least 22%, which shows that this method is better than the other five methods in automatically recognizing Weibo new emotional words.
【作者单位】：山西大学计算机与信息技术学院;计算智能与中文信息处理教育部重点实验室(山西大学);
【基金】：国家863计划项目(2015AA015407) 国家自然科学基金资助项目(61175067,61272095,61432011,61573231,U1435212) 山西省科技基础条件平台计划项目(2015091001-0102) 山西省回国留学人员科研项目(2013-014)~~
【分类号】：TP391.1

【参考文献】