One reason that is difficult to control today is, for instance, the mass dissemination of fake news and rumours. Conventional approaches in rumour detection are mainly based on text characteristics and, interestingly, ignore sentiment as an important parameter when categorizing rumours. This paper gives a preliminary overview of a dual-input convolutional neural network, which is sentiment-dependent and can enhance multi-class rumour detection. The proposed model processes two parallel inputs: One vector refers to feature engineering of semantic features and the other for sentiment analysis outcomes. This allows the model to process both the facts in the text and the overall sentiment intensity together with the separate markers to detect emotionally driven rumours more likely to be reposted. An experimental evaluation of a multi-class rumour detection task shows that the proposed model is more accurate and stable than traditional text-based techniques. This framework can handle textual information with sentiment analysis, and “Rumor Wall” provides a powerful solution for eliminating the propagation of rumours in different domains.