基于AT-LSTM的弹幕评论情感分析  

Emotional Analysis of Bullet-screen Comments Based on AT-LSTM

作　　者：庄须强 刘方爱[1,2] ZHUANG Xu-qiang;LIU Fang-ai(College of Information Science and Engineering,Shandong Normal University,Jinan Shandong 250014;Shandong Provincial Key Laboratory for Distributed Computer Software Novel Technology,Jinan Shandong 250014)

机构地区：[1]山东师范大学信息科学与工程学院,山东济南250014 [2]山东省分布式计算机软件新技术重点实验室,山东济南250014

出　　处：《数字技术与应用》2018年第2期210-212,共3页Digital Technology & Application

基　　金：国家自然科学基金(61572301);国家自然科学基金(90612003);山东省自然科学基金(ZR2013FM008)

摘　　要：弹幕评论能更准确、具体地反映出用户在观看视频时的即时情感和褒贬评价,因此本文提出了一种基于注意力机制的LSTM(AT-LSTM)情感分析模型。首先基于注意力机制更好的挖掘出整个弹幕评论中的情感关键词;然后利用LSTM模型有效结合视频中前后弹幕评论的情感依赖关系,最终提取出基于主题的"高光"视频片段。实验结果表明所提方法的准确度比传统LDA和LSTM方法有了进一步的提高。该模型可以帮助用户更准确的获取网络视频弹幕数据中包含的情感信息,进而提供了一种新的视频检索与视频推荐新途径。Bullet-screen comments can more accurately and specifically reflect the instant emotion and evaluation of users while watching the video,therefore,this paper proposed the AT-LSTM sentiment analysis model.First of all,through the attention mechanism,we can better dig out the emotional keywords in the whole bullet-screen comments;at the same time,LSTM model can more effectively combine the emotional dependency relationship between the front and rear bullet-screen comments in the video,and extract the theme based"highlight"video clips.The experimental results show that the accuracy of the proposed method is further improved compared with the traditional LDA and LSTM methods.The model can help users obtain the emotional information contained in the network video barrage data more accurately,and then provide a new way of video search and video recommendation.

关 键 词：深度学习 弹幕评论 情感分析 AT-LSTM 

分 类 号：G206[文化科学—传播学]