Research on Deep Extraction of Financial Entity Relations based on GAT Fusion Model

Jiayu Hong

doi:10.54691/gp0knb06

Authors

Jiayu Hong

DOI:

https://doi.org/10.54691/gp0knb06

Keywords:

Financial entities, relation extraction, graph attention, syntactic fusion.

Abstract

In response to the complexity of extracting entity relationships in Chinese financial texts, this study proposes a joint model (SynPOS-GAT) based on Graph Attention Network (GAT) that integrates dependency syntax and part-of-speech information. This model combines BERT for semantic encoding, utilizes BiLSTM to enhance the semantic representation of part-of-speech, and employs GAT to capture syntactic dependency structures, effectively addressing the problems of ambiguous entity boundaries and overlapping relationships in traditional methods. Experiments on the FinRE dataset show that the SynPOS-GAT model outperforms the comparison models in terms of accuracy, recall rate, and F1 value, with an F1 value of 44.78%, an average improvement of 10.08% compared to the baseline. Ablation experiments verified the crucial role of part-of-speech and syntactic information in the model performance. This study provides efficient technical support for the construction of financial knowledge graphs.

Downloads

Download data is not yet available.

References

[1] Shiraishi Y ,Kaneiwa K .A self-matching training method with annotation embedding models for ontology subsumption prediction[J].International Journal of Data Science and Analytics,2026,22(1):86-86.

[2] Nguyen T D H ,Le B .Faithful explanation of semantic role labelling with dependency and constituency feature importance[J].Neurocomputing,2026,676132996-132996.

[3] Suthakar G R A ,Sandesh J B .Enhanced image captioning with positional and dual attention using deep convolutional long short-term memory and emotional feedback mechanism[J].Multimedia Tools and Applications,2026,85(3):228-228.

[4] Baek W J ,Chung K .Scene graph optimization using hyperedge-based relationship extraction[J].Applied Soft Computing,2026,194114907-114907.

[5] Zhao Y ,Li Q ,Liu Y , et al.A multi-layer deep learning framework for joint extraction of geological disaster entities and relations[J].Geomatics, Natural Hazards and Risk,2025,16(1).

[6] Chen J K ,Chen Y ,Wang Z , et al. GMN-Zoomer: Learning graph similarity via hierarchical parsing, pooling and matching[J].Neural Networks,2026,200108778-108778.

[7] Xie Y ,Zheng X ,Xie S , et al. Fault prediction for aluminum electrolysis based on spatial-temporal attention dynamic graph convolutional networks[J].Control Engineering Practice,2026,172106878-106878.

[8] Zhou L ,Bai L .Unsupervised temporal knowledge graph entity alignment via BERT and structure adaptation[J].Expert Systems With Applications,2026,315131860-131860.

[9] Tong K ,Gan H Z ,Wang J S , et al.Prelimbic cortex to zona incerta pathway mediates recognition memory recovery delay through down-regulating astrocytic GAT-3[J].iScience,2026,29(3):115078-115078.

[10] Assis B V J ,Machado G H ,Rocha A Y , et al.A validated LLE–LTP–LC–MS/MS workflow for pesticide biomonitoring in human plasma: Application to occupational exposure assessment in rural workers[J].Environmental Chemistry and Ecotoxicology,2026,81342-1355.