Regarding the PURE baseline

[zeal2000]

Dear Author, Hi, I have a question that is confusing me, in the original work of PURE, the performance of Rel and Rel+ on SciERC dataset is 50.1 and 36.8 respectively, what is the metric you are using in GPT-RE that can reach 68.45, thanks!

[edzq]

@zeal2000 Basically, they are working on "Relation Classification" rather than "extraction". They defined their task in section 2.1 . In the PURE or PL-Marker paper, they are working on the end-to-end relation extraction, which has two steps, 1) extracting the entities and 2) classifying the potential relations of entity pairs.
The input of this paper is sentence + entity information.

[zeal2000]

@zeal2000 Basically, they are working on "Relation Classification" rather than "extraction". They defined their task in section 2.1 . In the PURE or PL-Marker paper, they are working on the end-to-end relation extraction, which has two steps, 1) extracting the entities and 2) classifying the potential relations of entity pairs. The input of this paper is sentence + entity information.

thank you! I see~