Addressing the issue of inadequate classification accuracy in named entity recognition via entity word-to-relationship modeling, we propose a method using rotational attention. Firstly, the text is encoded using the BERT model and Bi-LSTM, fol‐ lowed by extraction of features from input text using convolutional neural networks. Subsequently, the extracted feature sequence is inputted into the rotational attention model for output probability calculations, and the MLP layer is used for output classification. The study's outcomes affirm the efficacy and feasibility of the technique proposed in this paper, as it successfully yields superior re‐ sults on mainstream English databases including CADEC, GENIA and CoNLL2003, for named entity recognition exercise.