Metin commited on
Commit
f400ca2
·
verified ·
1 Parent(s): 0a66698

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +73 -3
README.md CHANGED
@@ -1,3 +1,73 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - text2text-generation
5
+ - question-answering
6
+ language:
7
+ - tr
8
+ size_categories:
9
+ - 1K<n<10K
10
+ ---
11
+
12
+ ![WikiRAG-TR Icon](docs/WikiRAG_TR.png)
13
+
14
+ # Dataset Summary
15
+ WikiRAG-TR is a dataset of 6K (5999) question and answer pairs which synthetically created from introduction part of Turkish Wikipedia Articles. The dataset is created to be used for Turkish Retrieval-Augmented Generation (RAG) tasks.
16
+
17
+ ## Dataset Information
18
+ - **Number of Instances**: 5999 (5725 synthetically generated question-answer pairs, 274 augmented negative samples)
19
+ - **Dataset Size**: 1.2 MB
20
+ - **Language**: Turkish
21
+ - **Dataset License**: apache-2.0
22
+ - **Dataset Category**: Text2Text Generation
23
+ - **Dataset Domain**: STEM and Social Sciences
24
+
25
+ ## WikiRAG-TR Pipeline
26
+
27
+ The creation of the dataset was accomplished in two main phases, each represented by a separate diagram.
28
+
29
+ ### Phase 1: Subcategory Collection
30
+ ![Subcategory collection diagram](docs/collecting_subcategories.png)
31
+
32
+ In this initial phase:
33
+ 1. A curated list of seed categories was decided, including science, technology, engineering, mathematics, physics, chemistry, biology, geology, meteorology, history, social sciences, and more.
34
+ 2. Using these seed categories, subcategories were recursively gathered from Wikipedia.
35
+ - **Recursion depth** was set to 3 and the **number of subcategories** to collect was limited to 100 for each depth layer.
36
+ 3. For each step, following subcategory types were filtered out:
37
+ - Subcategories containing **NSFW words**.
38
+ - Subcategories that only contain **lists of items**
39
+ - Subcategories used as **templates**
40
+ 4. Articles from the resulting subcategory list were acquired.
41
+
42
+ ### Phase 2: Dataset Generation
43
+ ![Creating WikiRAG-TR Diagram](docs/creating_wikirag_tr.png)
44
+
45
+ The second phase involved the following steps:
46
+ 1. Introduction sections were extracted from the articles gathered in Phase 1.
47
+ - If the introduction was **too short** or **too long** (less than 50 or more than 2500 characters), the article was discarded.
48
+ - If the introduction contained **NSFW words**, the article was discarded.
49
+ - If the introduction contained **equations**, the article was discarded.
50
+ - If the introduction section was **empty**, the article was discarded.
51
+ 2. The filtered introductions were fed into a large language model to generate synthetic question and answer pairs.
52
+ 3. For each resulting row in the dataset (containing an introduction, question, and answer), the following operations were performed:
53
+ - Unrelated contexts (introductions) were gathered from other rows to add false positive retrievals to the context.
54
+ - These unrelated contexts were appended to a list.
55
+ - The related context was added to this list. (In some cases, the relevant context was omitted to create **negative samples** where the answer indicates the model can't answer the question due to insufficient information. These negative samples were created separately, ensuring all original questions have corresponding answers.)
56
+ - The list was shuffled to **randomize the position** of the relevant context.
57
+ - The list elements were joined using the '\n' character.
58
+
59
+ ## Considerations for Using the Data
60
+ The generated answers are usually short and concise. This may lead to models trained on this dataset to generate short answers.
61
+
62
+ ## Dataset Columns
63
+ - `id`: Unique identifier for each row.
64
+ - `question`: The question generated by the model.
65
+ - `answer`: The answer generated by the model.
66
+ - `context`: The augmented context containing both relevant and irrelevant information.
67
+ - `is_negative_response`: Indicates whether the answer is a negative response (0: No, 1: Yes).
68
+ - `number_of_articles`: The number of article introductions used to create the context.
69
+
70
+ # Attributions
71
+ <a href="https://www.flaticon.com/free-icons/globe" title="globe icons">Globe icons created by Freepik - Flaticon</a>
72
+
73
+ <a href="https://www.flaticon.com/free-icons/search" title="search icons">Search icons created by Freepik - Flaticon</a>tle="search icons">Search icons created by Freepik - Flaticon</a>