Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -11,4 +11,71 @@ tags:
|
|
11 |
pretty_name: DiscoEval Benchmar
|
12 |
size_categories:
|
13 |
- 10K<n<100K
|
14 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
pretty_name: DiscoEval Benchmar
|
12 |
size_categories:
|
13 |
- 10K<n<100K
|
14 |
+
---
|
15 |
+
|
16 |
+
# DiscoEval Benchmark Datasets
|
17 |
+
|
18 |
+
## Overview
|
19 |
+
|
20 |
+
The DiscoEval benchmark datasets offer a collection of tasks designed to evaluate natural language understanding models in the context of discourse analysis and coherence. These tasks encompass various aspects of discourse comprehension and generation:
|
21 |
+
|
22 |
+
1. **Sentence Positioning**
|
23 |
+
- **Datasets**: Arxiv, Wikipedia, Rocstory
|
24 |
+
- **Description**: Determine the correct placement of a sentence within a given context or document.
|
25 |
+
|
26 |
+
2. **Binary Sentence Ordering**
|
27 |
+
- **Datasets**: Arxiv, Wikipedia, Rocstory
|
28 |
+
- **Description**: Choose between two possible orders for a pair of sentences, identifying the more coherent structure.
|
29 |
+
|
30 |
+
3. **Discourse Coherence**
|
31 |
+
- **Datasets**: Ubuntu IRC channel, Wikipedia
|
32 |
+
- **Description**: Evaluate the coherence of a discourse or conversation by determining if a sequence of sentences forms a coherent conversation.
|
33 |
+
|
34 |
+
4. **Sentence Section Prediction**
|
35 |
+
- **Dataset**: Constructed from PeerRead
|
36 |
+
- **Description**: Predict the section or category to which a sentence belongs within a scientific paper, based on the content and context.
|
37 |
+
|
38 |
+
5. **Discourse Relations**
|
39 |
+
- **Datasets**: RST Discourse Treebank, Penn Discourse Treebank
|
40 |
+
- **Description**: Identify and classify discourse relations between sentences or text segments, helping to reveal the structure and flow of discourse.
|
41 |
+
|
42 |
+
## Dataset Sources
|
43 |
+
|
44 |
+
- **Arxiv**: A repository of scientific papers and research articles.
|
45 |
+
- **Wikipedia**: An extensive online encyclopedia with articles on diverse topics.
|
46 |
+
- **Rocstory**: A dataset consisting of fictional stories.
|
47 |
+
- **Ubuntu IRC channel**: Conversational data extracted from the Ubuntu Internet Relay Chat (IRC) channel.
|
48 |
+
- **PeerRead**: A dataset of scientific papers frequently used for discourse-related tasks.
|
49 |
+
- **RST Discourse Treebank**: A dataset annotated with Rhetorical Structure Theory (RST) discourse relations.
|
50 |
+
- **Penn Discourse Treebank**: Another dataset with annotated discourse relations, facilitating the study of discourse structure.
|
51 |
+
|
52 |
+
## Use Cases
|
53 |
+
|
54 |
+
The DiscoEval benchmark datasets are valuable for training and evaluating natural language understanding models, particularly those focusing on discourse analysis, document structure prediction, and text generation tasks.
|
55 |
+
|
56 |
+
## Citation
|
57 |
+
|
58 |
+
Ensure you cite the appropriate sources and papers associated with each dataset when using them in your research or applications.
|
59 |
+
|
60 |
+
## License
|
61 |
+
|
62 |
+
Each dataset within the DiscoEval benchmark may have distinct licensing terms. Always review and comply with the individual dataset's licensing terms and conditions when using them.
|
63 |
+
|
64 |
+
## Loading Data Examples
|
65 |
+
|
66 |
+
### Loading Data for Sentence Positioning Task
|
67 |
+
|
68 |
+
```python
|
69 |
+
from datasets import load_dataset
|
70 |
+
|
71 |
+
# Load the Sentence Positioning dataset
|
72 |
+
dataset = load_dataset("disco_eval/sentence_positioning")
|
73 |
+
|
74 |
+
# Access the train, validation, and test splits
|
75 |
+
train_data = dataset["train"]
|
76 |
+
validation_data = dataset["validation"]
|
77 |
+
test_data = dataset["test"]
|
78 |
+
|
79 |
+
# Example usage: Print the first few training examples
|
80 |
+
for example in train_data[:5]:
|
81 |
+
print(example)
|