Rumor detection has been the focus of public opinion analysis and a topical area of in natural language processing research. Rumor detection which uses the deep learning method mainly extracts features such as rumor text, users information and communication structure, and determines rumors by inputting a suitable rumor detection model and using trained classifiers. In this work, a rumor detection model called BGLA is proposed, it merges the propagation structure and content features, the extracted feature vectors and adjacency matrix are passed through the BiGCN model to obtain the causal characteristics of rumor propagation along the relational chain from top to bottom, and the structural features obtained by bottom-up aggregation, the characteristics of the propagation and dispersion of rumors are simulated respectively. Then the BiLSTM model with an attention mechanism is added to obtain the temporal characteristics of the text, while allowing the model to focus more on the part of the input sequence that is relevant to the current prediction, and finally the event labels are calculated after the fully connected and softmax layers. Experiments on Twitter15 and Twitter16 show that the accuracy of the proposed method is improved by 1.2% and 1.9%, respectively, compared to the best baseline method.