Update README.md
Browse files
README.md
CHANGED
@@ -1,68 +1,64 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
# M-ABSA
|
5 |
|
6 |
-
This repo contains the data
|
7 |
|
8 |
[](https://arxiv.org/abs/2502.11824)
|
9 |
|
10 |
|
11 |
# Data Description:
|
|
|
12 |
|
13 |
All datasets are stored in the data/ folder:
|
14 |
|
15 |
-
- All dataset contains
|
16 |
|
17 |
```
|
18 |
domains = ["coursera", "hotel", "laptop", "restaurant", "phone", "sight", "food"]
|
19 |
```
|
20 |
-
- Each dataset contains
|
21 |
```
|
22 |
langs = ["ar", "da", "de", "en", "es", "fr", "hi", "hr", "id", "ja", "ko", "nl", "pt", "ru", "sk", "sv", "sw", "th", "tr", "vi", "zh"]
|
23 |
```
|
24 |
-
- Each dataset is divided into training, validation, and test sets. Each sentence is separated by __"####"__, with the first part being the sentence and the second part being the corresponding triplet. Here is an example, where the triplet includes __[aspect, category, sentiment]__.
|
25 |
|
26 |
-
|
27 |
-
This coffee brews up a nice medium roast with exotic floral and berry notes .####[['coffee', 'food quality', 'positive', 'nice']]
|
28 |
-
```
|
29 |
-
|
30 |
-
## Requirements
|
31 |
-
|
32 |
-
We recommend you to install the specified version of the following packages:
|
33 |
-
|
34 |
-
- transformers==4.0.0
|
35 |
-
- sentencepiece==0.1.91
|
36 |
-
- pytorch_lightning==0.8.1
|
37 |
-
|
38 |
-
## Quick Start for the Baseline
|
39 |
-
|
40 |
-
- Set up the environment as described in the above section.
|
41 |
-
- Download the pre-trained mT5-base model from [https://huggingface.co/google/mt5-base](https://huggingface.co/google/mt5-base) and place it under the folder mT5-base/ .
|
42 |
-
- Run command bash run.sh, which train the model on source language under UABSA/TASD task.
|
43 |
-
- Run command bash evaluate.sh, which test the model on target language under UABSA/TASD task.
|
44 |
-
|
45 |
-
****Detailed Usage****
|
46 |
-
|
47 |
-
We conduct experiments on two ABSA subtasks with M-ABSA dataset in the paper, you can change the parameters in run.sh to try them:
|
48 |
|
49 |
```
|
50 |
-
|
51 |
-
--dataset hotel \
|
52 |
-
--model_name_or_path mt5-base \
|
53 |
-
--paradigm extraction \
|
54 |
-
--n_gpu 0 \
|
55 |
-
--do_train \
|
56 |
-
--do_direct_eval \
|
57 |
-
--train_batch_size 16 \
|
58 |
-
--gradient_accumulation_steps 2 \
|
59 |
-
--eval_batch_size 16 \
|
60 |
-
--learning_rate 3e-4 \
|
61 |
-
--num_train_epochs 5
|
62 |
```
|
63 |
|
64 |
-
-
|
65 |
-
- $dataset refers to one of the seven datasets in [food, restaurant, coursera, laptop, sight, phone, hotel]
|
66 |
|
67 |
|
68 |
## Citation
|
@@ -78,5 +74,4 @@ If the code or dataset is used in your research, please star our repo and cite o
|
|
78 |
primaryClass={cs.CL},
|
79 |
url={https://arxiv.org/abs/2502.11824},
|
80 |
}
|
81 |
-
```
|
82 |
-
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
task_categories:
|
4 |
+
- token-classification
|
5 |
+
- text-classification
|
6 |
+
language:
|
7 |
+
- ar
|
8 |
+
- da
|
9 |
+
- de
|
10 |
+
- en
|
11 |
+
- es
|
12 |
+
- fr
|
13 |
+
- hi
|
14 |
+
- hr
|
15 |
+
- id
|
16 |
+
- ja
|
17 |
+
- ko
|
18 |
+
- nl
|
19 |
+
- pt
|
20 |
+
- ru
|
21 |
+
- sk
|
22 |
+
- sv
|
23 |
+
- sw
|
24 |
+
- th
|
25 |
+
- tr
|
26 |
+
- vi
|
27 |
+
- zh
|
28 |
+
tags:
|
29 |
+
- aspect-based-sentiment-analysis
|
30 |
+
size_categories:
|
31 |
+
- 100K<n<1M
|
32 |
+
---
|
33 |
# M-ABSA
|
34 |
|
35 |
+
This repo contains the data for our paper ****M-ABSA: A Multilingual Dataset for Aspect-Based Sentiment Analysis****.
|
36 |
|
37 |
[](https://arxiv.org/abs/2502.11824)
|
38 |
|
39 |
|
40 |
# Data Description:
|
41 |
+
This is a dataset suitable for the __multilingual ABSA__ task with __triplet extraction__.
|
42 |
|
43 |
All datasets are stored in the data/ folder:
|
44 |
|
45 |
+
- All dataset contains __7__ domains.
|
46 |
|
47 |
```
|
48 |
domains = ["coursera", "hotel", "laptop", "restaurant", "phone", "sight", "food"]
|
49 |
```
|
50 |
+
- Each dataset contains __21__ languages.
|
51 |
```
|
52 |
langs = ["ar", "da", "de", "en", "es", "fr", "hi", "hr", "id", "ja", "ko", "nl", "pt", "ru", "sk", "sv", "sw", "th", "tr", "vi", "zh"]
|
53 |
```
|
|
|
54 |
|
55 |
+
- The labels contain triplets with __[aspect term, aspect category, sentiment polarity]__. Each sentence is separated by __"####"__, with the first part being the sentence and the second part being the corresponding triplet. Here is an example, where the triplet includes __[aspect term, aspect category, sentiment polarity]__.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
```
|
58 |
+
This coffee brews up a nice medium roast with exotic floral and berry notes .####[['coffee', 'food quality', 'positive']]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
```
|
60 |
|
61 |
+
- Each dataset is divided into training, validation, and test sets.
|
|
|
62 |
|
63 |
|
64 |
## Citation
|
|
|
74 |
primaryClass={cs.CL},
|
75 |
url={https://arxiv.org/abs/2502.11824},
|
76 |
}
|
77 |
+
```
|
|