RUCAIBox / RecBole-TRM

MIT License
13 stars 5 forks source link

How to train NRMS model ? #2

Open vic4code opened 1 year ago

vic4code commented 1 year ago

I am glad to see this repo for news recommendation task! I recently tried to implement NRMS or other attention-based models on the MIND dataset, but found that the Dataset, Dataloader, Trainer, Recommender in the original design of RecBole needed to be modified to fit this specific task. Are you planning to integrate this repo back and rearrange your unified framework RecBole by task? If so, we can easily and conveniently run experiments on different datasets.

By the way, I tried to run the NMRS model with the following command.

python run_recbole_trm.py -m NRMS -t 8

The MIND dataset doesn't seem to be ready yet, some preprocessing is needed to get behaviors.tsv and npbehaviors.tsv defined in NRMS.yaml, how can I train the NRMS model correctly, thanks!

shamanez commented 1 year ago

Any update on this ?

wenqisun commented 1 year ago

Due to the use of text data in news recommendation, the original MIND dataset needs to be configured here. In the future, we will consider developing a separate news recommendation library based on the existing foundation. Thank you.