CC-News-En Support

**Dataset Information:**

A large (40M document) news corpus derived from CCNews, with associated query variations (UQVs).

Currently, the corpus is best used for efficiency work, since there are no "official" qrels from pooled systems.

**Links to Resources:**

- All data: https://go.unimelb.edu.au/u3nj (or https://cloudstor.aarnet.edu.au/plus/s/M8BvXxe6faLZ4uE)
- Documents: See the `warc` directory and the following (janky) script: https://github.com/JMMackenzie/CC-News-Tools/blob/master/get_warc.sh
- Queries: See `queries/cc-news-queries.json`
- Qrels: See `queries/cc-news-per-topic-qrels.txt`
- Overview paper: https://dl.acm.org/doi/10.1145/3340531.3412762

**Dataset ID(s):**

`cc-news-en` - Document corpus
`cc-news-en/queries` -- Queries and qrels

**Supported Entities**

- [x] docs
- [x] queries
- [x] qrels
- [] scoreddocs
- [] docpairs

**Additional comments/concerns/ideas/etc.**

- One problem is that the data is currently stored on AARNET and can be slow or difficult to download; there's almost 1TB of data here -- Perhaps supporting the full collection is not the best way forward right now.

- Note that the qrels are "pseudo qrels" since the UQV generation task was based on a re-finding simulation, so there is only one qrel per topic.

- In the queries file, you probably want to just use the `uid` and the `spellnorm_query` as the identifier/query pairs. Since there are multiple queries per topic, it's easiest to just re-map the identifiers when doing evaluation.

- We also provide crowdworker data and keystroke-level data, a pre-built [CIFF](https://github.com/osirrc/ciff) index from Anserini with pre-normalized queries, etc. But those are probably not useful here.

I'm happy to provide any support you may need in making the dataset more visible or accessible to the community, so let me know how I can help. Great project, by the way!