SPECTER: Document-level Representation Learning Using Citation-informed Transformers | Awesome LLM Papers

SPECTER: Document-level Representation Learning Using Citation-informed Transformers

Arman Cohan, Sergey Feldman, Iz Beltagy, Doug Downey, Daniel S. Weld Β· Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics Β· 2020

Representation learning is a critical ingredient for natural language processing systems. Recent Transformer language models like BERT learn powerful textual representations, but these models are targeted towards token- and sentence-level training objectives and do not leverage information on inter-document relatedness, which limits their document-level representation power. For applications on scientific documents, such as classification and recommendation, the embeddings power strong performance on end tasks. We propose SPECTER, a new method to generate document-level embedding of scientific documents based on pretraining a Transformer language model on a powerful signal of document-level relatedness: the citation graph. Unlike existing pretrained language models, SPECTER can be easily applied to downstream applications without task-specific fine-tuning. Additionally, to encourage further research on document-level models, we introduce SciDocs, a new evaluation benchmark consisting of seven document-level tasks ranging from citation prediction, to document classification and recommendation. We show that SPECTER outperforms a variety of competitive baselines on the benchmark.

Similar Work
Loading…