search

duynguyen-0203 / miner

MINER: Multi-Interest Matching Network for News Recommendation
23 stars 5 forks source link

Access of Google drive #1

Open quqxui opened 2 years ago

quqxui commented 2 years ago

Hi,

Thank you very much for the open-source codes, which has high readability. But i get a problem of the datasets. I have not access to your Google drive.

I have applied for permission. I'd appreciate it if you can open the access.

Also, i want to know whether it is the official code? And what is the combination of hyperparameters for best result of paper?

Best regards

duynguyen-0203 commented 2 years ago

Thanks for your issue.

This is just my implementation for the paper, not the official code.

Sorry that the data I tested is private. You can use the MIND dataset (https://msnews.github.io/) instead. You should replace some parameters in the confg/train.txt file, especially those about pretrained_tokenizer and pretrainend_embedding.

The hyperparameters I selected are based on paper. But like I said, this is not an official code, so I just try to code as close to paper as possible. I haven't tested on the MIND, so I'm not sure if I can reproduce the results as shown in the paper.

I hope the answer is useful to you. Thank you.

zztMermory commented 2 years ago

hello, thanks for sharing. I wonder that how the Miner's performance is on your own dataset.

dxxxmmm commented 8 months ago

你好，我想问一下应该怎么处理得到"category2id" and "user2id"这两个文件呢？