Concerning the problem of insufficient image feature extraction and ignorance of single-modal internal relations and the interactions between single-modal and multi-modal， a text and image information based Multi-Modal Deep Fusion （MMDF） model was proposed. Firstly， the Bi-Gated Recurrent Unit （Bi-GRU） was used to extract the rich semantic features of the text， and the multi-branch Convolutional-Recurrent Neural Network （CNN-RNN） was used to extract the multi-level features of the image. Then the inter-modal and intra-modal attention mechanisms were established to capture the high-level interaction between the fields of language and vision， and the multi-modal joint representation was obtained. Finally， the original representation of each modal and the fused multi-modal joint representation were re-fused according to their attention weights to strengthen the role of the original information. Compared with the Multimodal Variational AutoEncoder （MVAE） model， the proposed model has the accuracy improved by 1.9 percentage points and 2.4 percentage points on the China Computer Federation （CCF） competition and the Weibo datasets respectively. Experimental results show that the proposed model can fully fuse multi-modal information and effectively improve the accuracy of false information detection.

Under the constrains of very short message content on social media platforms， a large number of empty forwards in the transmission structure， and the mismatch between user roles and contents， a rumor detection model based on user attribute information and message content in the propagation network， namely GMB_GMU， was proposed. Firstly， user propagation network was constructed with user attributes as nodes and propagation chains as edges， and Graph Attention neTwork （GAT） was introduced to obtain an enhanced representation of user attributes； meanwhile， based on this user propagation network， the structural representation of users was obtained by using node2vec， and it was enhanced by using mutual attention mechanism. In addition， BERT （Bidirectional Encoder Representations from Transformers） was introduced to establish the source post content representation of the source post. Finally， to obtain the final message representation， Gated Multimodal Unit （GMU） was used to integrate the user attribute representation， structural representation and source post content representation. Experimental results show that the GMB_GMU model achieves an accuracy of 0.952 on publicly available Weibo data and can effectively identify rumor events， which is significantly better than the propagation algorithms based on Recurrent Neural Network （RNN） and other neural network benchmark models.