This paper proposes a hybrid deep model called CSI for fake news detection, which integrates three key characteristics of fake news: text, user response, and source. The model consists of three modules: Capture, Score, and Integrate. The Capture module uses a Recurrent Neural Network (LSTM) to capture the temporal pattern of user activity on articles, incorporating both textual and temporal features. The Score module extracts user representations and assigns scores based on their participation in group behavior. The Integrate module combines the outputs of the first two modules to classify articles as fake or not. Experimental results on real-world datasets show that CSI achieves higher accuracy than existing models and extracts meaningful latent representations of both users and articles. The model does not require a social graph, domain knowledge, or assumptions about the distribution of behaviors. CSI is general and can be applied to various datasets, capturing all three characteristics of fake news for accurate classification. The model also provides user scores that indicate suspicious behavior. The results demonstrate that CSI effectively captures and leverages the three characteristics of text, response, and source for accurate fake news detection.This paper proposes a hybrid deep model called CSI for fake news detection, which integrates three key characteristics of fake news: text, user response, and source. The model consists of three modules: Capture, Score, and Integrate. The Capture module uses a Recurrent Neural Network (LSTM) to capture the temporal pattern of user activity on articles, incorporating both textual and temporal features. The Score module extracts user representations and assigns scores based on their participation in group behavior. The Integrate module combines the outputs of the first two modules to classify articles as fake or not. Experimental results on real-world datasets show that CSI achieves higher accuracy than existing models and extracts meaningful latent representations of both users and articles. The model does not require a social graph, domain knowledge, or assumptions about the distribution of behaviors. CSI is general and can be applied to various datasets, capturing all three characteristics of fake news for accurate classification. The model also provides user scores that indicate suspicious behavior. The results demonstrate that CSI effectively captures and leverages the three characteristics of text, response, and source for accurate fake news detection.