Add README.md
Browse files
README.md
CHANGED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Finetuend `bert-base-multilignual-cased` model on Thai sequence and token classification datasets
|
2 |
+
|
3 |
+
<br>
|
4 |
+
|
5 |
+
Finetuned XLM Roberta BASE model on Thai sequence and token classification datasets
|
6 |
+
The script and documentation can be found at [this repository](https://github.com/vistec-AI/thai2transformers).
|
7 |
+
|
8 |
+
<br>
|
9 |
+
|
10 |
+
## Model description
|
11 |
+
|
12 |
+
<br>
|
13 |
+
|
14 |
+
We use the pretrained cross-lingual BERT model (mBERT) as proposed by [[Devlin et al., 2018]](https://arxiv.org/abs/1810.04805). We download the pretrained PyTorch model via HuggingFace's Model Hub (https://huggingface.co/bert-base-multilignual-cased)
|
15 |
+
<br>
|
16 |
+
|
17 |
+
|
18 |
+
|
19 |
+
## Intended uses & limitations
|
20 |
+
|
21 |
+
<br>
|
22 |
+
|
23 |
+
You can use the finetuned models for multiclass/multilabel text classification and token classification task.
|
24 |
+
|
25 |
+
<br>
|
26 |
+
|
27 |
+
**Multiclass text classification**
|
28 |
+
|
29 |
+
|
30 |
+
- `wisesight_sentiment`
|
31 |
+
|
32 |
+
4-class text classification task (`positive`, `neutral`, `negative`, and `question`) based on social media posts and tweets.
|
33 |
+
|
34 |
+
- `wongnai_reivews`
|
35 |
+
|
36 |
+
Users' review rating classification task (scale is ranging from 1 to 5)
|
37 |
+
|
38 |
+
- `generated_reviews_enth` : (`review_star` as label)
|
39 |
+
|
40 |
+
Generated users' review rating classification task (scale is ranging from 1 to 5).
|
41 |
+
|
42 |
+
**Multilabel text classification**
|
43 |
+
|
44 |
+
- `prachathai67k`
|
45 |
+
|
46 |
+
Thai topic classification with 12 labels based on news article corpus from prachathai.com. The detail is described in this [page](https://huggingface.co/datasets/prachathai67k).
|
47 |
+
|
48 |
+
|
49 |
+
**Token classification**
|
50 |
+
|
51 |
+
- `thainer`
|
52 |
+
|
53 |
+
Named-entity recognition tagging with 13 named-entities as descibed in this [page](https://huggingface.co/datasets/thainer).
|
54 |
+
|
55 |
+
- `lst20` : NER NER and POS tagging
|
56 |
+
|
57 |
+
Named-entity recognition tagging with 10 named-entities and Part-of-Speech tagging with 16 tags as descibed in this [page](https://huggingface.co/datasets/lst20).
|
58 |
+
|
59 |
+
<br>
|
60 |
+
|
61 |
+
## How to use
|
62 |
+
|
63 |
+
<br>
|
64 |
+
|
65 |
+
The example notebook demonstrating how to use finetuned model for inference can be found at this [Colab notebook](https://colab.research.google.com/drive/1Kbk6sBspZLwcnOE61adAQo30xxqOQ9ko)
|
66 |
+
|
67 |
+
<br>
|
68 |
+
|
69 |
+
|
70 |
+
**BibTeX entry and citation info**
|
71 |
+
|
72 |
+
```
|
73 |
+
@misc{lowphansirikul2021wangchanberta,
|
74 |
+
title={WangchanBERTa: Pretraining transformer-based Thai Language Models},
|
75 |
+
author={Lalita Lowphansirikul and Charin Polpanumas and Nawat Jantrakulchai and Sarana Nutanong},
|
76 |
+
year={2021},
|
77 |
+
eprint={2101.09635},
|
78 |
+
archivePrefix={arXiv},
|
79 |
+
primaryClass={cs.CL}
|
80 |
+
}
|
81 |
+
```
|