Question about DRG TF-IDF implementation

[iamkatelim]

Hello Mr Wu,

Background

I'm trying to reproduce the work in your paper "Mask and Infill: Applying Masked Language Model to Sentiment Transfer" for a different dataset. I've noticed that in the released code, there's a folder called drg_tf_idf which appears to be used for sentiment attribute markers.

Questions

1. Could you explain what "drg" stands for in drg_tf_idf?
2. Is this a modification of the fusion method described in Section 3.2, or something else entirely?
3. Are there any specific parameters or preprocessing steps needed to reproduce these scores for a new dataset?

My implementation attempt

I've attempted to implement the calculation by combining the frequency-ratio method with attention-based weighting as described in the paper, but I'd like to confirm if this is the correct approach.

Thank you in advance for any clarification you can provide!