根据IPCC（Intergovernmental Panel on Climate Change）的第六次评估报告（AR6：Assessment Report 6），极端降水事件将越来越频繁；内涝等现象给城市基础设施和经济带来的负面影响也愈发凸显（周波涛，2021）。自然灾害中相关者交换信息和看法，社交媒体凭借其及时性、双向互动性以及包含巨大信息的能力、极大地融入了公众的生活成为灾害中链接政府和公众的桥梁，通过社交媒体数据将公众的关注、情感、需求情况与风险沟通结合，分析社交媒体、政府如何在风险沟通方面发挥更好地的作用，达到防灾减灾的目的。

       本文研究造成巨大影响、引发公众高度关注的河南2021.7暴雨灾害事件，通过python爬取的7月15日至8月3日的98078条微博数据（其中包含官方媒体报道文本）、7月17日至7月23日的每小时降水量数据及灾情数据进行分析。首先通过文本分析法建立了河南暴雨洪涝灾害事件关键词分类表，根据本事件信息内容细分为26个类别，分析公众关注灾害信息的时空变化，并通过网络分析法分析公众对灾害影响与响应感知；其次通过SnowNLP情感分析模型判断公众的情绪极性，并进一步分析公众情感变化的时空差异及影响公众消极情绪感知的事件；最后通过文本分析将公众的应急救助需求细分为8个需求指标，并用TF-IDF（Term Frequency- Inverse Document Frequency）关键字统计分析方法计算公众和政府的救助需求指标值。由此本文从公众关注、公众情感、公众救援需求方面分析了公众感知情况，在此基础上，分析政府和媒体在本次事件中风险沟通情况，并给出建议，本文主要的结论是：

     （1）①根据公众信息关注情况河南暴雨洪涝灾害事件演化过程划分为潜伏期、爆发期、持续期、淡化期四个阶段，平均每日微博条数分别为220条/天、10672条/天、9608条/天、2569条/天。②暴雨洪涝自然灾害事件的微博文本主要呈现致灾因子、灾害影响、响应三个主题，整体上，公众最为关注的是响应信息（占比74.2%），其次是灾害影响信息（占比20.4%）。不同阶段公众对不同类型的灾害信息的关注热度不同，潜伏期公众最为关注山洪等致灾因子的预警信息；其他阶段内涝成为公众最为关注的致灾因子，人员伤亡和交通受阻为公众最为关注的灾害影响，灾民网上求救、捐款捐物和现场救援为公众最为关注的响应行为。灾区河南与非灾区的公众对灾害信息的关注并未呈现较大差异，对于暴雨、内涝、洪水、基础设施损毁影响以及灾民网上求助信息，灾区公众会更为关注。③公众重在感知影响与响应，本次事件出现人员伤亡、交通受阻、经济损失、能源通讯、生产生活、次生灾害6个凝聚子群，其中人员伤亡、现场救援、建筑倒塌、交通受阻、通信受阻作为本次暴雨洪涝灾害的核心事件，与其他节点联系紧密。

    （2）①本次灾害事件中，公众的积极情感占主导，爆发期（7月20-21日），消极情绪的占比较高，为40.91%。灾区河南与非灾区公众消极情绪表达的微博条数的变化趋势是基本一致的，从占比情况看，灾区河南的消极情绪微博条数占比较高。②在爆发期，公众在线表达消极情感异常活跃，两天间“彻夜不眠”地表达消极情绪。而在持续期，公众的消极情绪表达呈现规律的状态，9时、13时、20时，公众较为活跃，夜间2时至7时，公众表达减少，至7月25日夜间基本无消极情绪表达。③以人民日报及河南日报为代表的官方媒体对暴雨灾害事件的报道详尽，贯穿事件始终，在此期间，官方媒体通过更新救援情况、赞扬救援行为、发布灾后重建信息进行着风险沟通，抚慰公众情绪。

     （3）公众的救助需求可以分为灾后紧急救援需求、基本生活保障需求、公共基础设施需求三个方面，可进一步细分为8个类别。公众的救助需求呈现动态性，对人员搜救需求最高，其次是道路交通保障需求。公众的救助需求基本得到了政府的回应，7月20日，公众的道路交通保障需求最高，政府对该需求的回应也最高，7月21-25日，公众的人员搜救需求最高，政府对该需求的回应也最高。

       由此，进行灾害风险沟通时，政府媒体应把握好舆论引导的时、度、效，抓住公众关注度高（事件爆发期）的时期，及时公开透明灾情及救援进展，进行辟谣，以每日8时、11时、17时、20时等公众关注较为活跃的时间为宜。在持续期，注重识别负面信息，跟进事件，纾解公众消极情绪。消退期注重反思、安抚，稳定社会秩序。另外，政府主流媒体在灾害中要进行积极的情感引导，有目的地传递信息，安抚、渲染公众情感，如赞扬志愿者、鼓励公众团结等。政府也可以有效地利用社交媒体，了解并回应公众相应的需求，在灾害爆发初期，可多关注公众的基础设施维护需求，出现人员伤亡后，要及时回应公众的人员搜救需求。

       综上，本文以“河南2021.7暴雨灾害事件”为例，通过社交媒体数据研究公众感知并提出了风险沟通的建议。研究特色在于构建了本事件的关键词表，分析公众关注的时空变化；构建了本事件的情感语料库了解公众情感的时空变化，不足在于语料库的普适性有待验证，并且精度还有提升的空间；构建了应急救助需求评价指标体系，分析公众的应急救助需求和政府的需求回应。本研究创新性在于基于社交媒体数据，从关注、情感、需求三方面分析公众感知，从公众、媒体、政府角度进行风险沟通研究，并提出了针对性的建议。

    According to the sixth assessment report (AR6:Assessment Report 6) of IPCC (Intergovernmental Panel on Climate Change), extreme precipitation events will become more and more frequent; The negative impact of waterlogging on urban infrastructure and economy has become more and more prominent. In natural disasters, stakeholders exchange information and opinions. Social media, with its timeliness, two-way interaction, and the ability to contain amounts of information, has greatly integrated into the public's life and become a bridge linking the government and the public in disasters. Through social media data, the public's concerns, sentiment, rescue demand, and risk communication are combined, and how social media and the government can play a better role in risk communication is analyzed to achieve the purpose of disaster prevention and mitigation.

    In this paper, the rainstorm flood disaster of July 2021 in Henan Province, which caused great impact and aroused great public concern, was studied. Through python crawling 98,078 pieces of Weibo data from July 15th to August 3rd (including official media reports), hourly precipitation data from July 17th to July 23rd, and disaster data, it was analyzed. Firstly, a keyword classification table of rainstorm and flood disaster events in Henan Province is established by text analysis, which is subdivided into 26 categories according to the information content of this event, and the temporal and spatial changes of public concern about disaster information are analyzed, and the public's perception of disaster impact and response are analyzed by network analysis. Secondly, judge the public's emotional polarity through the SnowNLP emotional analysis model, and further analyze the temporal and spatial differences of public emotional changes and the events that affect negative emotional perception; Finally, the public's emergency rescue demand is subdivided into eight demand indicators through text analysis, and the index values of public and government rescue demand are calculated by using TF-IDF (Term Frequency- Inverse Document Frequency) keyword statistical analysis method. Therefore, this paper analyzes the public perception from the aspects of public concern, public sentiment, and public rescue demand. On this basis, it analyzes the risk communication between the government and the media in this incident and gives suggestions. The main conclusions are as follows:

   (1)① According to the public information concern, the evolution process of the rainstorm flood disasters in Henan Province can be divided into four stages: incubation period, outbreak period, duration period, and desalination period. The average daily number of Weibo is 220/day, 10672/day, 9608/day, and 2569/day respectively. ② The Weibo text of rainstorm flood natural disasters mainly presents three themes: disaster-causing factors, disaster impact, and response. On the whole, the public is most concerned about response information (accounting for 74.2%), followed by disaster impact information (accounting for 20.4%). The public pays different attention to different types of disaster information at different stages, and the public is most concerned about the early warning information of disaster-causing factors such as flash floods in the incubation period; At other stages, waterlogging has become the most concerned disaster-causing factor for the public. Casualties and traffic jams are the most concerned disasters, and online help, donations, and on-site rescue are the most concerned response behaviors for the public. There is no big difference in the public's attention to disaster information between the disaster-stricken areas and the non-disaster areas in Henan Province. The public in disaster-stricken areas will pay more attention to the impact of heavy rain, waterlogging, floods, and damage to infrastructure and online help information for victims. ③ The public focuses on the perception of influence and response. There are six cohesive subgroups in this incident, including casualties, traffic jams, economic loss, energy communication, production and life, and secondary disasters. Among them, casualties, on-site rescue, building collapse, traffic jams, and communication jams are the core events of this rainstorm and flood disaster, which are closely related to other nodes.

    (2)① In this disaster event, the public's positive emotions dominated, and during the outbreak period (July 20-21), the proportion of negative emotions was relatively high, accounting for 40.91%. The changing trend of the number of negative emotions expressed by the public in disaster-stricken areas and non-disaster areas in Henan Province is the same, from the perspective of the proportion, the number of negative emotions expressed by Weibo in the disaster-stricken areas in Henan Province is relatively high. In terms of proportion, the number of negative emotions expressed by Weibo in disaster-stricken areas in Henan Province accounts for about 30%, the number of negative emotions expressed by Weibo in non-disaster areas in Henan Province accounts for about 70%. ② During the outbreak period, the public was very active in expressing negative emotions online, and they stayed up all night for two days to express negative emotions. In the duration, the public's negative emotional expression showed a regular state. At 9: 00, 13: 00, and 20: 00, the public was more active, and from 2: 00 to 7: 00 at night, the public's expression decreased, and there was no negative emotional expression until the night of July 25. (3) The official media, represented by People's Daily and Henan Daily, reported the rainstorm flood disaster in detail throughout the incident. During this period, the official media conducted risk communication by updating the rescue situation, praising the rescue behavior, and releasing post-disaster reconstruction information to soothe public sentiment.

    (3) The public's rescue demands can be divided into three aspects: post-disaster emergency rescue demands, basic living security demands, and public infrastructure demands, which can be further subdivided into eight categories. The public's demand for rescue is dynamic, and the demand for personnel search and rescue is the highest, followed by the demand for road traffic security. The public's demand for rescue has been responded to by the government. On July 20th, the public's demand for road traffic security was the highest, and the government's response to this demand was the highest. From July 21st to July 25th, the public's demand for personnel search and rescue was the highest, and the government's response to this demand was also the highest.

    Therefore, when conducting risk communication, the government media should grasp the time, degree and effect of public opinion guidance, seize the period of high public concern (the outbreak of the incident), disclose the disaster situation and rescue progress in a timely manner, and dispel rumors. It is appropriate to take 8: 00, 11: 00, 17: 00 and 20: 00 every day when public concern is more active. In the continuous period, pay attention to identify negative information, follow up the incident, and relieve the public's negative emotions. In the fading period, we should pay attention to reflection and appeasement, and stabilize social order. In addition, the mainstream media of the government should give positive emotional guidance in disasters, transmit information purposefully, appease and render public feelings, such as praising volunteers and encouraging public unity. The government can also effectively use social media to understand and respond to the corresponding needs of the public. In the early days of the disaster, it can pay more attention to the public's infrastructure maintenance needs, and respond to the public's personnel search and rescue needs in time after casualties occur.

    To sum up, this paper takes the "2021.7 rainstorm disaster in Henan" as an example, studies public perception through social media data and puts forward suggestions on risk communication. The research feature is to construct the keyword list of this event, and analyze the temporal and spatial changes of public concern; The emotional corpus of this event is constructed to understand the temporal and spatial changes of public emotions. The shortcomings are that the universality of the corpus needs to be verified, and there is still room for improvement in accuracy. The evaluation index system of emergency rescue demand is constructed, and the public emergency rescue demand and the government's demand response are analyzed. The innovation of this study lies in analyzing public perception from three aspects: concern, emotion and demand, conducting risk communication research from the perspectives of the public, the media and the government, and putting forward targeted suggestions.