Abstract: The recognition of negative sentiment intensity in financial public opinion on social media is crucial for pre-venting financial risks and maintaining market stability. With the diversification of online expression formats, multimodal information combining text and images has become the primary way for the public to convey financial sentiments and opinions. To fully leverage the correlation and complementarity among different modalities and improve recognition performance, this study proposes a multimodal financial public opinion negative sentiment intensity recognition model based on GCN-Stacking collaborative enhancement. First, the model adopts a multi-model consistency annotation strategy to obtain high-confidence sentiment la-bels and uses the CLIP model to extract basic multimodal features of text and images. Subsequently, a multi-modal graph structure is constructed, and Graph Convolutional Network (GCN) is employed to realize structural association embedding and enhancement of features, making up for the defect that traditional multimodal features ignore potential inter-modal correlations. Finally, a Stacking framework is built through base classifier selection and two-layer ensemble modeling to optimize the performance of sentiment intensity recognition. Experimental results show that under the optimal model configuration, the F1-score reaches 83.57% and the AUC reaches 95.05%. Compared with the optimal single-text model, optimal single-image model, and original CLIP multimodal model, the F1-score is improved by 2.48%, 5.27%, and 1.87% respectively. These results fully verify the effectiveness of the proposed method, which can provide technical support and d