annieske commited on
Commit
2190ebd
1 Parent(s): 38557f0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -3
README.md CHANGED
@@ -11,9 +11,12 @@ size_categories:
11
  ### Dataset Summary
12
 
13
  The Turku WebQA dataset is a Finnish Question-Answer dataset that has been extracted from different CommonCrawl sources (Parsebank, mC4-Fi, CC-Fi).
14
- It has 237,000 question-answer pairs (altogether 290,000 questions, but not all have an answer). The questions with no answers can be discarded by taking out the rows with None (null).
 
15
  The codebase as well as the raw data can be found on [GitHub](https://github.com/TurkuNLP/register-qa).
16
 
 
 
17
  ### Data Fields
18
 
19
  - `source`: a `string` feature. Tells whether the question-answer pair is extracted from Parsebank, mC4-Fi or CC-Fi.
@@ -21,7 +24,7 @@ The codebase as well as the raw data can be found on [GitHub](https://github.com
21
  - `question`: a `string` feature.
22
  - `answer`: a `string` feature. Can also be None (null).
23
 
24
- ### Manual evalution of the pairs
25
 
26
  To get an idea on how good the extracted pairs were, a sample was annotated for noisy artefacts, insufficient answers and missing context.
27
  The evaluation showed that there is variation between the different source corpora.
@@ -34,7 +37,6 @@ The evaluation showed that there is variation between the different source corpo
34
  | Parsebank (N=22) | 0,23 | 0,14 | 0,07 |
35
 
36
 
37
-
38
  ### Citing
39
 
40
  Citing information coming soon!
 
11
  ### Dataset Summary
12
 
13
  The Turku WebQA dataset is a Finnish Question-Answer dataset that has been extracted from different CommonCrawl sources (Parsebank, mC4-Fi, CC-Fi).
14
+
15
+ The dataset has 237,000 question-answer pairs (altogether 290,000 questions, but not all have an answer). The questions with no answers can be discarded by taking out the rows with None (null).
16
  The codebase as well as the raw data can be found on [GitHub](https://github.com/TurkuNLP/register-qa).
17
 
18
+ The extracted question-answer pairs include various topics from the source corpora, some of which are explored in the paper for which the citing information can be found below.
19
+
20
  ### Data Fields
21
 
22
  - `source`: a `string` feature. Tells whether the question-answer pair is extracted from Parsebank, mC4-Fi or CC-Fi.
 
24
  - `question`: a `string` feature.
25
  - `answer`: a `string` feature. Can also be None (null).
26
 
27
+ ### Manual Evalution of the Pairs
28
 
29
  To get an idea on how good the extracted pairs were, a sample was annotated for noisy artefacts, insufficient answers and missing context.
30
  The evaluation showed that there is variation between the different source corpora.
 
37
  | Parsebank (N=22) | 0,23 | 0,14 | 0,07 |
38
 
39
 
 
40
  ### Citing
41
 
42
  Citing information coming soon!