ZHAO Shuxu1， LIU Lijiao1， MA Qinjing2
（1. School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China；
2. School of Information Engineering, Gansu Forestry Polytechnic, Tianshui 741020, China）
Abstract： With the development of short video industry, video and bullet screen have become important ways to spread public opinions. Public attitudes can be timely obtained through emotional analysis on bullet screen, which can also reduce difficulties in management of online public opinions. A convolutional neural network model based on multi-head attention is proposed to solve the problem of how to effectively model relations among words and identify key words in emotion classification tasks with short text contents and lack of complete context information. Firstly, encode word positions so that order information of input sequences can be used by the model. Secondly, use a multi-head attention mechanism to obtain semantic expressions in different subspaces, effectively capture internal relevance and enhance dependent relationships among words, as well as highlight emotional weights of key emotional words. Then a dilated convolution is used to increase the receptive field and extract more features. On this basis, the above multi-attention mechanism is combined with a convolutional neural network to model and analyze the seven emotional categories of bullet screens. Testing from perspectives of model and dataset, experimental results can validate effectiveness of our approach. Finally, emotions of bullet screens are visualized to provide data supports for hot event controls and other fields.
Key words： bullet screen； text sentiment classification； self-attention mechanism； visual analysis; hot events controlReferences
［1］Yang M G, Wang L Y, Wan H L. An analysis of the new form of audience interaction in the network live broadcasting environment—taking bilibili video barrage network as an example. Design, 2017, 11(21): 24-25.
［2］Xia M X. Comments polarity classification based on emotion analysis and movie recommender system design and implementation. Hefei: Anhui University, 2016: 1-7.
［3］Lin Y F. Research and implementation of feature selection in Chinese text classification. Xi’an: Xidian University, 2014: 1-21.
［4］Mou L, Li G, Liu Y, et al. Building program vector representations for deep learning. In: Proceedings of International Conference on Knowledge Science Engineering and Management, Sibiu, Lomnia. 2014.
［5］Kim Y. Convolutional neural networks for sentence classification. Eprint Arxiv, 2014: 1-5.
［6］Chua, Leon O. CNN: a paradigm for complexity. World scientific, Republic of Singapore, 1998: 24-178.
［7］Gers F A, Schmidhuber J, Cummins F, et al. Learning to forget: continual prediction with LSTM. Neural Computation, 1999, 12(10): 2451-2471.
［8］Kalchbrenner N, Grefenstette E, Blunsom P. A convolutional neural network for modelling sentences. Eprint Arxiv, 2014.
［9］Santos N D, Gattit M. Deep convolutional neural networks for sentiment analysis of short texts. In: Proceedings of International Conference on Computational Linguistics, Dublin, Ireland. 2014: 69-78.
［10］Zheng Y Y, Xu J, Xiao Z. Application of sentiment analysis and visualization methods in network video barrage data analysis. Modern Library and Information Technology, 2015(11): 82-90.
［11］Deng Y, Zhang C X, Li J F. Video clip recommendation model based on barrage sentiment analysis. Computer Applications, 2017, 37(4): 1065-1070.
［12］Hong Q, Wang S Y, Zhao Q P, et al. Video user group classification based on barrage comments sentiment analysis and clustering algorithms. Computer Engineering and Science, 2018, 40(6): 1125-1139.
［13］Zhuang X Q, Liu F G. AT-LSTM-based sentiment analysis of barrage reviews. Digital Technology and Application, 2018, 332(2): 220-222.
［14］Wang X Y. Research and application of sentiment classification of network video barrage based on image analysis. Beijing: Beijing University of Posts and Telecommunications, 2018.
［15］Zhang Y S, Zheng J, Huang G J, et al. Microblog sentiment analysis method based on a double attention model. Journal of Tsinghua University (Natural Science Edition), 2018, 58(2): 122-130.
［16］Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need. Eprint Arxiv, 2017: 1-15.
［17］Zhu Y B. Research on sentiment classification method based on deep learning and self-attention mechanism. Xi’an: Xi’an University of Technology, 2019.
［18］Ahmadzadeh R. Random search algorithm. Malaysia: UniMAP, 2012: 13-53.
［19］Liu Y, Xing Y Y. Research and application of text classification based on improved random forest algorithm. Journal of Computer Systems, 2019, 28(5): 222-227.
基于自注意力机制的弹幕文本情绪分类模型
赵庶旭1， 刘李姣1， 马秦靖2
（1. 兰州交通大学 电子与信息工程学院， 甘肃 兰州 730070； 2. 甘肃林业职业技术学院 信息工程学院， 甘肃 天水 741020）
摘要:随着短视频行业的发展， 视频及弹幕已然成为舆论传播的重要途径， 对弹幕进行情绪分析能及时获取民众态度， 解决网络舆情管理的困难。 弹幕文本内容短、 几乎不具有完整上下文信息的特性， 而较好地对词语间关系进行建模并有效识别关键词语对情绪分类任务而言是极其重要的问题， 本文提出一种基于多头注意力的卷积神经网络模型。 首先， 通过对词语位置进行编码使模型能够利用输入序列的顺序信息。 其次， 利用多头注意力机制获取不同子空间的语义表达， 有效捕获词语内部的相关性， 增强词语间依存关系， 突出关键情感词的情感权重， 并利用膨胀卷积增大感受野， 提取更多特征， 在此基础上将多头注意力机制与卷积神经网络融合对弹幕文本的七个情绪类别进行建模分析。 从模型及数据集角度对比实验， 结果表明本文模型能有效提高情绪分类精准度。 最后将弹幕情感进行可视化分析， 以期为热点事件管控等领域起到一定的数据支撑作用。
关键词:弹幕文本; 文本情绪分类; 自注意力机制； 可视化分析； 热点事件管控
引用格式：ZHAO Shuxu， LIU Lijiao， MA Qinjing. Sentiment classification model for bullet screen based on self-attention mechanism. Journal of Measurement Science and Instrumentation, 2021, 12（4）： 479-488. DOI： 10.3969／j.issn.1674-8042.2021.04.012
[full text view]
