[目的/意义] 在"新冠"疫情这类突发公共卫生事件中，网络社交媒体上迅速产生大量关于疫情的言论，其中包含不少蓄意传播的谣言，不仅危害公众心理健康，而且会影响应对公共卫生事件的方案实施。识别突发公共卫生事件的谣言能够使民众正确面对危机，为社会安定、网络治理起到积极的维护作用。[方法/过程] 首先对采集到的疫情期间已被证实的谣言进行深度分析，提取谣言文本的主要特征，包括上下文特征、话题类别特征、情感程度特征、关键词特征等；然后针对文本分类模型中的文本特征表达较为单一的问题，利用不同的模型对提取的谣言文本特征进行向量化，并对各类文本特征进行加强和融合。其中通过TF-IDF计算的词向量权重在捕获上下文特征的同时，能够加强词粒度的关键词特征信息。最后，使用BiLSTM+DNN模型对融合的特征向量进行分类判别。[结果/结论] 实验结果表明，话题类别、情感程度等特征对谣言识别均有贡献，特别是经过强化后的词向量与其他特征融合后对识别准确率有明显提升，召回率、F1值等指标均达到90%以上，效果超过其他的谣言识别模型，说明笔者所构建的方法能够很好地实现对突发公共卫生事件背景下的谣言识别。
[Purpose/significance] In public health emergencies such as the COVID-19 epidemic, a large number of statements about the epidemic have quickly been generated on social media on the Internet, including many rumors that endanger public mental health and affect the implementation of national policies. Detecting these remarks and identifying the rumors can enable the people to respond to public health emergencies correctly, and play a positive role in maintaining social stability and network governance.[Method/process] Firstly, the confirmed rumors during the epidemic were collected for in-depth analysis, and the main features of the rumor text were extracted, including context features, topic category features, sentiment level features, keyword features, etc.; then aiming at the problem that the text feature expression in the text classification model was relatively single, different models were used to vectorize the extracted rumor text features, and then a rumor recognition model based on multi-feature fusion was constructed. In the construction of this model, TF-IDF was used to strengthen the word vector, so that the word vector can merge the keyword feature information of the word granularity while capturing the context feature. Finally, this paper used the BiLSTM+DNN model to classify the fused feature vectors.[Result/conclusion] The experimental results show that features such as topic category and emotional level all contribute to the recognition of rumors, especially the fusion of the strengthened word vector and other features to significantly improve the recognition accuracy, recall rate, F1 measure, etc. The indicators all reached more than 90%, and the effect surpassed other rumor recognition models, indicating that the method constructed in this article can respond well to the task of rumor recognition in the context of public health emergencies.
[1] 国务院政策网站.突发公共卫生事件应急条例[EB/OL].[2021-05-29]. http://www.gov.cn/zhengce/2020-12/26/content_5574586.htm.
[2] 张丽,毛浩然.突发公共事件网络谣言治理研究综述[J].南京晓庄学院学报,2020,36(3):49-55,122.
[3] CARLOS C, MARCELO M, BARBARA P. Information credibility on twitter[C]//Proceedings of the 20th international conference on World Wide Web. New York:ACM, 2011:675-684.
[4] YANG F, YU X, LIU Y, et al. Automatic detection of rumor on Sina Weibo[C]//Proceedings of the ACM SIGKDD Workshop on mining data semantics. New York:ACM, 2012:1-7.
[5] 贺刚,吕学强,李卓,等.微博谣言识别研究[J].图书情报工作,2013,57(23):114-120.
[6] 夏松,林荣蓉,刘勘.网络谣言敏感词库的构建研究-以新浪微博谣言为例[J].知识管理论坛,2019,4(5):267-275.
[7] 李钢,王聿达.基于受众画像的新型耦合社交网络谣言传播模型研究[J].现代情报,2020,40(1):123-133,143.
[8] KWON S, CHA M, JUNG K, et al. Prominent features of rumor propagation in online social media[C]//2013 IEEE 13th international conference on data mining. Dallas:IEEE, 2013:1103-1108.
[9] MA J, GAO W, WEI Z, et al. Detect rumors using time series of social context information on Microblogging Websites[C]//ACM international conference on information and knowledge management. New York:ACM, 2015:1751-1754.
[10] 王志宏,过弋.微博谣言事件自动检测研究[J].中文信息学报,2019,33(6):132-140.
[11] KOTTETI M, DONG X, QIAN L. Multiple time-series data analysis for rumor detection on social media[C]//2018 IEEE international conference on big data. Seattle:IEEE, 2018:4413-4419.
[12] MA J, GAO W, MITRA P, et al. Detecting rumors from microblogs with recurrent neural networks[C]//Proceedings of the 25th international joint conference on artificial intelligence. New York:AAAI Press, 2016:3818-3824.
[13] LI L, CAI G, CHEN N. A rumor events detection method based on deep bidirectional GRU neural network[C]//IEEE international conference on image. Piscataway:IEEE, 2018:755-759.
[14] 王星宇. 基于长短期记忆网络及深层特征的谣言识别系统[D].保定:河北大学,2020.
[15] 刘勘, 杜好宸. 基于深度迁移网络的Twitter谣言检测研究[J]. 数据分析与知识发现. 2019, 3(10):47-55.
[16] SIVA G, DEEPAK P, CHENG L, et al. Unsupervised fake news detection:a graph-based approach[C]//Proceedings of the 31st ACM conference on hypertext and social media. Florida:ACM, 2020:75-83.
[17] MARRA F, GRAGNAIELLO D, COZZOLINO D, et al. Detection of GAN-generated fake images over social networks[C]//Proceedings of 2018 IEEE conference on multimedia information processing and retrieval. Piscataway:IEEE, 2018:384-389.
[18] QIAN F, GONG C, SHARMA K, et al. Neural user response generator:fake news detection with collective user intelligence[C]//Proceedings of the 27th international joint conference on artificial intelligence. Stockholm:IJCAI, 2018:3834-3840.
[19] 樊荣. 社会化媒体中突发公共事件谣言传播特征及传播行为预测[D].大连:东北财经大学,2019.
[20] 曾子明,王婧.基于LDA和随机森林的微博谣言识别研究——以2016年雾霾谣言为例[J].情报学报,2019,38(1):89-96.
[21] 王林,王可,吴江.社交媒体中突发公共卫生事件舆情传播与演变——以2018年疫苗事件为例[J].数据分析与知识发现,2019,3(4):42-52.
[22] 李丽华,韩思宁.暴恐事件网络舆情传播机制及预防研究——英国典型案例的实证分析[J].情报杂志,2019,38(11):102-111,54.
[23] 曹娟,张勇东,李锦涛,等.一种基于密度的自适应最优LDA模型选择方法[J].计算机学报,2008(10):1780-1787.
