Abstract: Aiming at the problems of long content, complex syntax structure, and the inability of traditional language models to express polysemous words using static word vectors, this paper proposes an open source intelligence text classification algorithm based on the fusion of BERT and RCNN. BERT was connected to the RCNN model as an embedded layer. The context vector information was captured through a recurrent convolution layer to learn the global semantic relationship of the input text. The maximum feature extraction was performed through the pooling layer. The Softmax classifier was used to classify the extracted features. The experimental results show that the classification accuracy of BERT-RCNN fusion model on the test set is higher than that of other mainstream models, which verifies the effectiveness of this algorithm in improving the classification effect of open source intelligence texts.