网络评论作为社交媒体平台的核心组成部分，已深刻重塑了人们的信息获取与意见表达模式。网络评论不仅为公众提供了一个思想碰撞、观点交流的场所，还在反映民众心态、塑造个体观点方面发挥着举足轻重的作用。然而，由于对大量评论编码的困难，既往的研究多依赖于问卷或仿真手段来探究表达意愿，缺乏与真实媒体环境及意见表达形式的紧密结合。因此，本研究旨在利用实际环境中的数据和人工智能技术，深入探究当前意见氛围对后续意见发表的影响，以及网络评论不同表现形式在这一过程中所起的作用。 本研究采用网络爬虫技术，对网易新闻中的热点新闻评论数据进行持续、定期的采集，以精准捕捉包括点赞和点踩在内的意见走向。随后，本研究运用大语言模型对意见进行编码，以实现意见倾向的精确划分。最后，借助机器学习中的随机森林算法，以及SHAP和ALE解释算法，对数据进行拟合与深入解析。 研究结果显示，在网易新闻评论区中，当用户感知到的意见气候不利于自己时，用户会倾向于减少以点赞和评论为形式的意见表达。具体而言，评论表达会因代表性的反对意见的出现而减少表达，而点赞表达则会在大部分评论均对用户不利时减少。值得注意的是，点踩表达难以受到不利意见气候的影响，反而在意见气候越不利时，点踩表达越为明显。此外，点踩意见表达形成的意见气候会产生社会支持，进而促进除点赞外其他意见形式的表达，从而有效阻止沉默螺旋的进一步发生。 综上所述，本研究表明，在网易新闻评论区中，意见气候对后续意见产生重要影响，且点击意见起到了重要作用，大部分群体压力以及社会支持的产生与其有紧密联系。因此，本研究建议在新闻评论区的设计中应重视点击意见的作用，以避免意见表达走向单一化，从而确保言论的多样性和健康性。

Online comments, as a core component of social media platforms, have profoundly reshaped the modes of information acquisition and opinion expression. Online comments not only provide a space for the collision of thoughts and exchange of viewpoints but also play a crucial role in reflecting public sentiment and shaping individual opinions. However, due to the difficulty of coding large volumes of comments, previous research has mostly relied on questionnaires or simulation methods to explore willingness to express opinions, lacking close integration with real media environments and forms of opinion expression. Therefore, this study aims to use data from actual environments and artificial intelligence technology to deeply explore the impact of the current opinion climate on subsequent opinion expression and the role of different forms of online comments in this process.  This study employs web crawling technology to continuously and periodically collect data on hot news comments from NetEase News to accurately capture opinion trends, including likes and dislikes. Subsequently, the study uses a large language model to encode opinions to achieve precise classification of opinion tendencies. Finally, with the help of the random forest algorithm in machine learning, as well as SHAP and ALE explanation algorithms, the data is fitted and deeply analyzed.  The results show that in the NetEase News comment section, when users perceive the opinion climate to be unfavorable to them, they tend to reduce their opinion expression in the form of likes and comments. Specifically, comment expression decreases with the emergence of representative opposing opinions, while likes decrease when most comments are unfavorable to the user. Notably, dislike expression is less influenced by an unfavorable opinion climate and becomes more pronounced when the opinion climate is more unfavorable. Furthermore, the opinion climate formed by dislike expressions generates social support, which in turn promotes the expression of II ABSTRACT other forms of opinions, except likes, effectively preventing the further occurrence of the spiral of silence. In summary, this study shows that in the NetEase News comment section, the opinion climate has a significant impact on subsequent opinions, and click opinions play an important role. The generation of group pressure and social support is closely related to them. Therefore, this study suggests that the design of news comment sections should emphasize the role of click opinions to avoid the homogenization of opinion expression, ensuring the diversity and healthiness of discourse.