Nested Entity Recognition of Integrates Crossposition Encoding and Feature Fusion: Nested Entity Recognition

In order to address the problems of low feature adhesion, fuzzy boundaries, and failure to effectively utilize multisentence sequences in the nested named entity recognition method, a nested entity recognition method that fuses cross-attention and feature drop-up is proposed. First, position information is added to characters through the designed cross-position encoding at the word embedding layer to strengthen boundary representation. Second, the proposed Drop-up is used to confuse features to enhance feature interaction. Then, a multi-layer entity recognition framework is designed based on the pointer network recognition to determine the beginning and end indexes of entities. Finally, a Pro-Clash method is designed to solve the conflict between entities and comprehensively improve generalization and recognition performance. To verify the effectiveness of the verification code model, tests were conducted on four datasets: Resume, Weibo, Ecommerce, and CNERTA. The F1 values were 96.85 %, 72.06 %, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 9. 1 5 \%}$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 4. 2 5 \%}$</tex>, respectively, which were <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 6 2 \%}, \mathbf{2. 5 4 \%}$</tex>, 0.6 %, and 0.03 % higher than the baseline model. The results show that the proposed method is generally superior to current mainstream methods and has good recognition results.