Currently, some researchers execute annotation operation for the chapter logical structure of the document content and pay attention to describe the logical structure and data content, eliminate various redundant style information, which can be conduce to processing document information more quickly and conveniently. The traditional annotation methods cannot identify the English streaming document structure effectively due to the rare corpus and complex corpus annotation. Therefore, this paper proposes a new logical structure annotation in English streaming document based on deep learning. First, the logical structure annotation system of English streaming document is established. Then, a three-stage logical structure annotation of semi-automatic English streaming document is proposed. In the first stage, separate annotation of English document metadata is realized by semi-automatic approach. In the second stage, the logical structure is automatically rebuilt. In the third stage, the eigenvector is automatically filled. The experiment results show that the proposed method can save labor cost and improve the accuracy and recall rate of the annotation compared with other state-of-the-art methods.