rebel
The REBEL project converts Relation Extraction into a seq2seq task, using BART-based autoregressive models for efficient extraction of relation triplets. By turning these triplets into linear sequences, it overcomes traditional pipeline limitations, and supports over 200 relation types. Integration possibilities with platforms like Hugging Face and spaCy enable easy adoption in various applications, achieving top performance in multiple benchmarks. The recent addition of mREBEL enhances multilingual extraction capabilities, covering a wide range of language datasets.
