This repository provides the generated document IDs (DocIDs) used for training and evaluating the DDRO (Direct Document Relevance Optimization) models.
Two types of DocIDs are included:
PQ (Product Quantization) DocIDs: Compact semantic representations based on quantized document embeddings.
TU (Title + URL) DocIDs: Tokenized document identifiers constructed from document titles and/or URLs.
pq_msmarco_docids.txt: PQ DocIDs for MS… See the full description on the dataset page:
https://huggingface.co/datasets/kiyam/ddro-docids.
