File size: 3,793 Bytes
b33e8bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
af9ab9f
 
 
 
 
b33e8bb
 
 
 
 
af9ab9f
 
eccb68a
 
 
 
 
1a6abba
 
a1b44ee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1a6abba
 
a1b44ee
d31727f
1a6abba
8c27518
1a6abba
555e0e5
1a6abba
 
 
d31727f
 
 
 
 
1a6abba
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
---
dataset_info:
  features:
  - name: qid
    dtype: string
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: context
    sequence: string
  - name: answers
    sequence: string
  - name: competition
    dtype: string
  - name: timestamp
    dtype: string
  - name: section
    dtype: string
  - name: number
    dtype: string
  - name: original_question
    dtype: string
  - name: original_answer
    dtype: string
  - name: original_additional_info
    dtype: string
  splits:
  - name: train
    num_bytes: 7981391
    num_examples: 2939
  - name: validation
    num_bytes: 2671680
    num_examples: 980
  download_size: 6275956
  dataset_size: 10653071
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
license: cc-by-sa-4.0
task_categories:
- question-answering
language:
- ja
---


## ⚠️ ! 注意 ! 

公開したデータセットは、私の不注意で本来学習させる用途には相応しくない、JAQKET テスト用のデータセットも train に含んでしまっています。
そのため、以下のように再度テスト用のデータを分離してからの利用をお勧めします。

```
import datasets

ds = datasets.load_dataset("hotchpotch/jaqket_v1_qa_wikija_context")
train_ds = ds["train"]  # type: ignore
valid_ds = ds["validation"]  # type: ignore

# 大元の "hotchpotch/jaqket_v1_qa_wikija_context" の train / valid 分割方法がよくないので、再度分割する

train_df = train_ds.to_pandas()
valid_df = valid_ds.to_pandas()
df = pd.concat([train_df, valid_df])

valid_target_section_names = ['リーダーボードテスト問題', 'ライブコンペテスト問題']
valid_df = df[df.section.isin(valid_target_section_names)]
train_df = df[~df.section.isin(valid_target_section_names)]
print(len(train_df), len(valid_df))
```


# jaqket_v1_qa_wikija_context


[AI王 公式配布データセット(JAQKET)](https://sites.google.com/view/project-aio/dataset?authuser=0)へ、Wikipedia のコンテキストを追加したデータセットです。

JAQKET で公開されており、かつライセンスが[CC BY-SA 4.0 DEED](https://creativecommons.org/licenses/by-sa/4.0/deed.ja)のデータをもとに、質問の回答が含まれる Wikipedia の passage をコンテキスト追加し HuggingFace Dataset を作成した物です。Q&A タスクにコンテキスト入れて学習させたいときや、質問 + コンテキストから回答をきちんと抽出できるのか(RAGのreader用途)、といったタスクの利用が想定されます。

データセットの context には、wikipedia の関連 passage の Top-3 (質問を特徴文ベクトルへ変換したもの類似passage Top-3)が格納されていて、このTop-3のうちのどこかの文字列に answer が含まれています。Top-3にanswerが含まれなかったデータは、取り除いています。データはtrain 2939件、validation 980件に split しています。

wikipedia passage には [singletongue/wikipedia-utils の passages-c400-jawiki-20230403](https://huggingface.co/datasets/singletongue/wikipedia-utils/viewer/passages-c400-jawiki-20230403)を、質問から関連する passage の取得には [hotchpotch/wikipedia-passages-jawiki-embeddings](https://huggingface.co/datasets/hotchpotch/wikipedia-passages-jawiki-embeddings)の `multilingual-e5-large-query` を使って類似ベクトル検索し抽出しています。


## データセットの作成スクリプト

- https://huggingface.co/datasets/hotchpotch/jaqket_v1_qa_wikija_context/blob/main/create_dataset.ipynb

## ライセンス

- [CC BY-SA 4.0 DEED](https://creativecommons.org/licenses/by-sa/4.0/deed.ja)