直播网站在线弹幕情感分析

Sentiment Analysis of Online Bullet Screen of Live Broadcast Websites

视频直播是近年来最火的新型产业。由于弹幕独特的在线实时、语言简略与互联网化特征,现有方法难以直接用于弹幕情感分析。为解决弹幕文本分析精度问题,针对弹幕语料库缺乏以及语言简略与互联网化特征,构建弹幕专属情感词典。针对直播弹幕语言的特性,提出一种基于改进SVM的情感分析模型。通过引入分类处理因子与梯度下降因子,降低了预测器的泛化误差。在此基础上,提出词向量、情感词、否定词和标点符号等多种融合特征的方法,将融合结果映射到向量空间上,再通过分类器进行情感分类。实验结果表明,改进的SVM分类器模型比未改进模型在精确率、召回率、F1值上分别提高3.8%、2.3%、1.1%。

Live video broadcasting is the hottest emerging industry in recent years.Due to the unique online real-time,simple language and Internet-based characteristics of the bullet screen,the existing methods are difficult to directly use the bullet screen′s emotional analysis.In order to solve the problem of the accuracy of the analysis of the bullet screen text,a sentiment analysis model based on an improved SVM is proposed here for the characteristics of the language of the live broadcast bullet screen.On this basis,a method of fusing various features such as word vectors,sentiment words,negation words and punctuation is proposed,and the fusion results are mapped onto the vector space,and then the sentiment is classified by a classifier.The experimental results show that the improved SVM classifier model proposed is 3.8%,2.3%and 1.1%higher than the evaluation indexes(accuracy,recall and F1 value)of the unimproved model,respectively.