multinews_sparse_oracle

This is a copy of the [Multi-News](https://huggingface.co/datasets/multi_news) dataset, except the input source documents of its `test` split have been replaced by a __sparse__ retriever. The retrieval pipeline used:

- __query__: The `summary` field of each example
- __corpus__: The union of all documents in the `train`, `validation` and `test` splits
- __retriever__: BM25 via [PyTerrier](https://pyterrier.readthedocs.io/en/latest/) with default settings
- __top-k strategy__: `"oracle"`, i.e. the number of documents retrieved, `k`, is set as the original number of input documents for each example

Retrieval results on the `test` set:

| Recall@100 | Rprec | Precision@k | Recall@k |
| ----------- | ----------- | ----------- | ----------- |
| 0.8775 | 0.7480 | 0.7480 | 0.7480 |

本数据集为Multi-News（https://huggingface.co/datasets/multi_news）的复刻版本，仅其`test`划分的输入源文档已替换为**稀疏检索器（sparse retriever）**的检索结果。所使用的检索流程如下：
- **查询（query）**：每个样本的`summary`字段
- **语料库（corpus）**：`train`、`validation`与`test`划分下的全部文档的并集
- **检索器（retriever）**：基于[PyTerrier](https://pyterrier.readthedocs.io/en/latest/)实现的BM25算法，采用默认配置
- **Top-k策略（top-k strategy）**：采用“神谕（oracle）”模式，即检索得到的文档数`k`设置为每个样本原始输入文档的数量

测试集上的检索性能指标如下：
| 召回率@100 | R精度（R-prec） | 精确率@k | 召回率@k |
| ----------- | ----------- | ----------- | ----------- |
| 0.8775 | 0.7480 | 0.7480 | 0.7480 |