Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,107 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
---
|
4 |
+
|
5 |
+
|
6 |
+
# Dataset Overview
|
7 |
+
|
8 |
+
This repository contains benchmark datasets for LLM-based topic discovery and traditional topic models. Original [GitHub](https://github.com/ahoho/topics?tab=readme-ov-file#download-data)
|
9 |
+
|
10 |
+
## [Bills Dataset](https://huggingface.co/datasets/zli12321/Bills)
|
11 |
+
|
12 |
+
The Bills Dataset is a collection of legislative documents with 32,661 bill summaries (train) from the 110th–114th U.S. Congresses, categorized into 21 top-level and 112 secondary-level topics.
|
13 |
+
|
14 |
+
|
15 |
+
- **Train Split**: 32.7K summaries.
|
16 |
+
- **Test Split**: 15.2K summaries.
|
17 |
+
|
18 |
+
### Loading the Bills Dataset
|
19 |
+
```
|
20 |
+
from datasets import load_dataset
|
21 |
+
|
22 |
+
# Load the train and test splits
|
23 |
+
train_dataset = load_dataset('zli12321/Bills', split='train')
|
24 |
+
test_dataset = load_dataset('zli12321/Bills', split='test')
|
25 |
+
```
|
26 |
+
|
27 |
+
## [Wiki Dataset](https://huggingface.co/datasets/zli12321/Wiki)
|
28 |
+
|
29 |
+
The Wiki dataset consists of 14,290 articles spanning 15 high-level and 45 mid-level topics, including widely recognized public topics such as music and anime.
|
30 |
+
|
31 |
+
- **Train Split**: 14.3K summaries.
|
32 |
+
- **Test Split**: 8.02K summaries.
|
33 |
+
|
34 |
+
## Synthetic Science Fiction (Pending internal clearance process)
|
35 |
+
|
36 |
+
Please cite us if you find the data and the papers useful, and do not hesitate to create an issue or email us if you have problems!
|
37 |
+
|
38 |
+
If you find LLM-based topic generation has hallucination or instability, and coherence not applicable to LLM-based topic models:
|
39 |
+
```
|
40 |
+
@misc{li2025largelanguagemodelsstruggle,
|
41 |
+
title={Large Language Models Struggle to Describe the Haystack without Human Help: Human-in-the-loop Evaluation of LLMs},
|
42 |
+
author={Zongxia Li and Lorena Calvo-Bartolomé and Alexander Hoyle and Paiheng Xu and Alden Dima and Juan Francisco Fung and Jordan Boyd-Graber},
|
43 |
+
year={2025},
|
44 |
+
eprint={2502.14748},
|
45 |
+
archivePrefix={arXiv},
|
46 |
+
primaryClass={cs.CL},
|
47 |
+
url={https://arxiv.org/abs/2502.14748},
|
48 |
+
}
|
49 |
+
```
|
50 |
+
|
51 |
+
If you use the human annotations or preprocessing:
|
52 |
+
```
|
53 |
+
@inproceedings{li-etal-2024-improving,
|
54 |
+
title = "Improving the {TENOR} of Labeling: Re-evaluating Topic Models for Content Analysis",
|
55 |
+
author = "Li, Zongxia and
|
56 |
+
Mao, Andrew and
|
57 |
+
Stephens, Daniel and
|
58 |
+
Goel, Pranav and
|
59 |
+
Walpole, Emily and
|
60 |
+
Dima, Alden and
|
61 |
+
Fung, Juan and
|
62 |
+
Boyd-Graber, Jordan",
|
63 |
+
editor = "Graham, Yvette and
|
64 |
+
Purver, Matthew",
|
65 |
+
booktitle = "Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)",
|
66 |
+
month = mar,
|
67 |
+
year = "2024",
|
68 |
+
address = "St. Julian{'}s, Malta",
|
69 |
+
publisher = "Association for Computational Linguistics",
|
70 |
+
url = "https://aclanthology.org/2024.eacl-long.51/",
|
71 |
+
pages = "840--859"
|
72 |
+
}
|
73 |
+
```
|
74 |
+
|
75 |
+
If you want to use the claim coherence does not generalize to neural topic models:
|
76 |
+
```
|
77 |
+
@inproceedings{hoyle-etal-2021-automated,
|
78 |
+
title = "Is Automated Topic Evaluation Broken? The Incoherence of Coherence",
|
79 |
+
author = "Hoyle, Alexander Miserlis and
|
80 |
+
Goel, Pranav and
|
81 |
+
Hian-Cheong, Andrew and
|
82 |
+
Peskov, Denis and
|
83 |
+
Boyd-Graber, Jordan and
|
84 |
+
Resnik, Philip",
|
85 |
+
booktitle = "Advances in Neural Information Processing Systems",
|
86 |
+
year = "2021",
|
87 |
+
url = "https://arxiv.org/abs/2107.02173",
|
88 |
+
}
|
89 |
+
```
|
90 |
+
|
91 |
+
|
92 |
+
If you evaluate ground-truth evaluations or stability:
|
93 |
+
```
|
94 |
+
@inproceedings{hoyle-etal-2022-neural,
|
95 |
+
title = "Are Neural Topic Models Broken?",
|
96 |
+
author = "Hoyle, Alexander Miserlis and
|
97 |
+
Goel, Pranav and
|
98 |
+
Sarkar, Rupak and
|
99 |
+
Resnik, Philip",
|
100 |
+
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2022",
|
101 |
+
year = "2022",
|
102 |
+
publisher = "Association for Computational Linguistics",
|
103 |
+
url = "https://aclanthology.org/2022.findings-emnlp.390",
|
104 |
+
doi = "10.18653/v1/2022.findings-emnlp.390",
|
105 |
+
pages = "5321--5344",
|
106 |
+
}
|
107 |
+
```
|