近年来，社交媒体上的计算宣传行为愈演愈烈，社交机器人被广泛用于操控舆论、制造话题、转移注意力，成为社交媒体操纵 的工具。准确识别社交机器人、分析社交机器人行为模式并了解社交机器人操控舆论的模式对社交机器人的治理至关重要。本 研究选取2019‐2020年11个热议的社会公共事件，从计算宣传的技术属性和社会属性入手，综合使用人工特征提取与深度学习方 法对220万微博账号的行为特征进行提取，基于模型融合等方法建立社交机器人识别模型。在此基础上，本研究根据每个事件中 的机器人占比，抽样出一个由14万微博用户构成的数据集，并建立逻辑回归模型。研究发现社交机器人识别算法的平均AUC得 分为0.88。微博上社会公共事件中的机器人既聪明又傻瓜。一方面，社交机器人借助低发文率、低活跃度逃避平台管制；另一方 面，它们又喜好在同一时间段群体行动，同时发布高同质性文案。从操纵策略上看主要采用扩音方式（72.2%）而非引导方式；从实 际效果来看，扩音作用也显著强于引导作用。综上，目前微博上的社交机器人在热点社会事件中表现出明显的特征和较高的可预 测性，其社交媒体操纵策略主要是扩大声量而非引导舆论。

In recent years, computational propaganda on social media has become increasingly fierce, and social bots have been widely used to manipulate public opinion, create topics and divert attention, becoming a tool for social media manipulation. Therefore, it is important to detect them and analyze the patterns of their behaviors. From both a technical and social perspective of computational propaganda, this study extracts both artificial and deep‑learning features of 2.2 million Weibo accounts to establish a social bots detection model for 11 topical issues in 2019‑2020. By using several methods such as dataset split and model integration, the average AUC of the model is over 0.88. Controlling the proportion of bots in each event, the study samples 140, 000 Weibo users for logistic regression and finds that the bots in social topical issues are both "smart" and "stupid". On the one hand, they conceal themselves with low activity. On the other hand, they prefer collective actions and highly homogeneous texts. They prefer to playing an amplifying role rather than a leading role, and their impact on opinion amplification is stronger than their impact on opinion leading. To sum up, social bots on microblog show obvious characteristics and high predictability in hot social events, and their strategy is mainly to expand the voice rather than to lead public opinion.