@article{fdi:010094282,
  title = {{GRU}-{SCANET}: unleashing the power of {GRU}-based sinusoidal capture network for precision-driven named entity recognition},
  author = {{H}appi {H}appi, {B}ill gates and {P}elap, {G}. {F}. and {S}ymeonidou, {D}. and {L}armande, {P}ierre},
  editor = {},
  language = {{ENG}},
  abstract = {{M}otivation {P}re-trained {L}anguage {M}odels ({PLM}s) have achieved remarkable performance across various natural language processing tasks. {H}owever, they encounter challenges in biomedical named entity recognition ({NER}), such as high computational costs and the need for complex fine-tuning. {T}hese limitations hinder the efficient recognition of biological entities, especially within specialized corpora. {T}o address these issues, we introduce {GRU}-{SCANET} ({G}ated {R}ecurrent {U}nit-based {S}inusoidal {C}apture {N}etwork), a novel architecture that directly models the relationship between input tokens and entity classes. {O}ur approach offers a computationally efficient alternative for extracting biological entities by capturing contextual dependencies within biomedical texts.{R}esults {GRU}-{SCANET} combines positional encoding, bidirectional {GRU}s ({B}i{GRU}s), an attention-based encoder, and a conditional random field ({CRF}) decoder to achieve high precision in entity labeling. {T}his design effectively mitigates the challenges posed by unbalanced data across multiple corpora. {O}ur model consistently outperforms leading benchmarks, achieving better performance than {B}io{BERT} (8/8 evaluations), {P}ub{M}ed{BERT} (5/5 evaluations), and the previous state-of-the-art ({SOTA}) models (8/8 evaluations), including {B}ern2 (5/5 evaluations). {T}hese results highlight the strength of our approach in capturing token-entity relationships more effectively than existing methods, advancing the state of biomedical{NER}.{A}vailability and implementation https://github.com/{ANR}-{DIG}-{AI}/{GRU}-{SCANET}.},
  keywords = {},
  booktitle = {},
  journal = {{B}ioinformatics {A}dvances},
  volume = {5},
  numero = {1},
  pages = {vbaf096 [10 p.]},
  year = {2025},
  DOI = {10.1093/bioadv/vbaf096},
  URL = {https://www.documentation.ird.fr/hor/fdi:010094282},
}