Au sujet de ce document
- sl:arxiv_author :
- sl:arxiv_firstAuthor : Arman Cohan
- sl:arxiv_num : 2004.07180
- sl:arxiv_published : 2020-04-15T16:05:51Z
- sl:arxiv_summary : 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.@en
- sl:arxiv_title : SPECTER: Document-level Representation Learning using Citation-informed Transformers@en
- sl:arxiv_updated : 2020-05-20T17:39:52Z
- sl:bookmarkOf : https://arxiv.org/abs/2004.07180
- sl:creationDate : 2022-01-29
- sl:creationTime : 2022-01-29T15:18:20Z
Documents with similar tags (experimental)
Préférences...