基于账户特征开展社交机器人检测，构建了一种用于识别恶意社交机器人账号的深度学习分类模型。该模型由五个层组成：一个包含10个神经元的输入全连接层、两个各含128个神经元的全连接层、一个Dropout层，以及一个输出全连接层。模型训练过程中，采用了多种激活函数，并结合Adam优化器进行优化。通过与其他四种基于机器学习的模型进行对比实验，验证了所提模型的有效性。本文深度学习模型在 F1 值方面优于其他模型，且准确率达到了次高水平。值得一提的是，基于随机森林构建的社交机器人识别模型在准确率等指标上也优于其他主流机器学习方法，展现出良好的性能。综上所述，深度学习技术在社交机器人识别的实验中表现出卓越的性能，能够满足实际研究的需求，可应用于社交平台机器人账号检测的实际场景

We carried out social bot detection based on account features and constructed a deep learning classification model for identifying malicious social bot accounts. The model consisted of five layers: an input fully connected layer containing 10 neurons, two fully connected layers containing 128 neurons each, a dropout layer, and an output fully connected layer. Multiple activation functions were used during model training and optimized in conjunction with the Adam optimizer. The effectiveness of the proposed model was verified by comparison experiments with four other machine learning based models. The deep learning model proposed in this experiment outperforms other models in terms of F1 value and reaches the next highest level of accuracy. It is worth mentioning that the social robot recognition model constructed based on random forest also outperforms other mainstream machine learning methods in terms of accuracy and other metrics, showing good performance. In summary, the deep learning technique shows excellent performance in the experiments of social robot recognition, which can meet the needs of practical research and can be applied to practical scenarios of robot account detection on social platforms