Some questions, about the downstream Author Name Disambiguation task, on the who is who dataset

[biandh]

We tried to reproduce oagbert's downstream task — Author Name Disambiguation, but the results we reproduced were far from given in the paper

In the paper, the expression is ”apply the embeddings generated by pre-trained models to solve name disambiguation from scratch“ . And the results of Unsupervised are also shown in the table 1, so I tried to use OAG-BERT-v2 for verification, But the effect is very poor, when only use title, Macro Pairewise F1 scores < 0.2

At the same time，the paper open sourced another version of OAG-BERT-sim on github, I used this model for verification, and found that the results were consistent with the results given in the paper table 1, which makes I'm confused, according to github, OAG-BERT-sim is a supervised fine-tune task.

So, how can I reproduce the results in Table 1 ?

@Somefive