Fanjin Zhang

Thanks for your reply. I modify the nonoverlapping.py file and run on the cora dataset. The loss is decreasing but nmi and modularity are not increasing. I implement _forward_ function...

sigmoid has been incorporated into the loss function ` cost = norm * F.binary_cross_entropy_with_logits(preds, labels, pos_weight=pos_weight) `

name_to_pubs_train contains matchings of persons and papers, which is to train global metric learning model and cluster size estimation model. name_to_pubs_test is for evaluation. Please see our paper for details.

内层字典的key是person id，value是这个人发表的论文id列表。论文id， 如XXX-1表示这个作者是第几作者，从零开始计数。 name_to_pubs_train_500.json: This file can be used for training data, which includes name-person-paper mapping relations. Data schema: This file is a dictionary (denoted as dic1) saved as a...

yes. The disambiguation results are obtained by clustering (in train.py).

It seems that the link works well here. You can try and download via [onedrive link](https://1drv.ms/u/s!AjyjU4F_oXtllmRV9aFPN1bpkEBY) alternatively.