Update README
Browse files
README.md
CHANGED
|
@@ -9,7 +9,7 @@ tags:
|
|
| 9 |
- travel
|
| 10 |
---
|
| 11 |
|
| 12 |
-
# AI21
|
| 13 |
|
| 14 |
The **AI21-Hotels** and **AI21-WorldCup** datasets were created to support research on *aggregative question answering* in open-book settings.
|
| 15 |
Aggregative questions require retrieving information from a large set of documents and applying reasoning over the collected text snippets.
|
|
@@ -20,8 +20,8 @@ These datasets were first introduced in the paper “Structured RAG for Answerin
|
|
| 20 |
## AI21-Hotels dataset
|
| 21 |
|
| 22 |
The **AI21-Hotels** dataset was automatically generated, with booking-like hotel pages, paired with aggregative queries.
|
| 23 |
-
The train set (in folder
|
| 24 |
-
The evaluation set (in folder
|
| 25 |
|
| 26 |
## AI21-WorldCup dataset
|
| 27 |
The **AI21-WorldCup** dataset consists of 22 Wikipedia pages, each corresponding to one WorldCup tournament held between 1930 and 2022,
|
|
|
|
| 9 |
- travel
|
| 10 |
---
|
| 11 |
|
| 12 |
+
# AI21-Hotels and AI2-WorldCup datasets
|
| 13 |
|
| 14 |
The **AI21-Hotels** and **AI21-WorldCup** datasets were created to support research on *aggregative question answering* in open-book settings.
|
| 15 |
Aggregative questions require retrieving information from a large set of documents and applying reasoning over the collected text snippets.
|
|
|
|
| 20 |
## AI21-Hotels dataset
|
| 21 |
|
| 22 |
The **AI21-Hotels** dataset was automatically generated, with booking-like hotel pages, paired with aggregative queries.
|
| 23 |
+
The train set (in folder AI21-Hotels-train) includes 50 pages, and 138 queries.
|
| 24 |
+
The evaluation set (in folder AI21-Hotels-eval) includes 350 pages, and 193 queries.
|
| 25 |
|
| 26 |
## AI21-WorldCup dataset
|
| 27 |
The **AI21-WorldCup** dataset consists of 22 Wikipedia pages, each corresponding to one WorldCup tournament held between 1930 and 2022,
|