Named entity recognition in water conservancy field based on multi-feature fusion and boundary perception

Named entity recognition in the water conservancy field is a crucial step in extracting water conservancy information, playing a key role in constructing water conservancy knowledge graphs and enhancing water conservancy knowledge question answering systems. However, existing methods still face challenges when dealing with text data in the field, including semantic diversity of entities, ambiguity in entity boundaries and insufficient feature utilization, resulting in limited recognition performance. To address these issues, this paper proposes a Named Entity Recognition Model based on Multi-feature Fusion and Boundary Perception (MFBP-NER). The model extracts multi-source features from three dimensions namely characters, pinyin, and dictionaries to obtain a more semantically rich feature representation. On this basis, an Adaptive Gating Fusion (AGF) mechanism is proposed to achieve dynamic weighting and effective fusion among multiple features, fully exploiting the deep semantic interaction relationships across different feature dimensions. Meanwhile, a BiGRU network is employed to further capture context dependencies and long-distance relationships within sequences. Additionally, a boundary perception module is incorporated into the global pointer framework to model semantic dependency relationships between spans, thereby enhancing entity boundary differences and enabling more accurately determine entity boundary. To evaluate the performance of MFBP-NER, experiments are conducted on three datasets with varying granularity. The results show that the proposed model achieves higher F1 scores compared to baseline models, demonstrating the effectiveness of the proposed approach in named entity recognition within the water conservancy field. It provides a viable technical foundation for constructing water conservancy knowledge graphs and developing question answering systems.