michal-stefanik
commited on
Commit
•
39e4834
1
Parent(s):
a3496e5
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
---
|
4 |
+
# Model Card for XLM-Roberta-large-reflective-conf4
|
5 |
+
|
6 |
+
This is a reflectivity classification model trained to distinguish different types of reflectivity in the reports of teaching students.
|
7 |
+
|
8 |
+
It was evaluated in a cross-lingual settings and was found to work well also in languages outside English -- see the results in the referenced paper.
|
9 |
+
|
10 |
+
## Model Details
|
11 |
+
|
12 |
+
- **Repository:** https://github.com/EduMUNI/reflection-classification
|
13 |
+
- **Paper:** https://link.springer.com/article/10.1007/s10639-022-11254-7
|
14 |
+
|
15 |
+
- **Developed by:** Jan Nehyba & Michal Stefanik, Masaryk University
|
16 |
+
- **Model type:** Roberta-large
|
17 |
+
- **Finetuned from model:** [XLM-R-large](https://huggingface.co/xlm-roberta-large)
|
18 |
+
|
19 |
+
## Usage
|
20 |
+
|
21 |
+
To match the training format, it is best to use the prepared wrapper that will format the classified sentence and its surrounding context in the expected format:
|
22 |
+
|
23 |
+
```python
|
24 |
+
from transformers import AutoConfig, AutoModelForSequenceClassification, AutoTokenizer
|
25 |
+
|
26 |
+
LABELS = ["Other", "Belief", "Perspective", "Feeling", "Experience",
|
27 |
+
"Reflection", "Difficulty", "Intention", "Learning"]
|
28 |
+
|
29 |
+
class NeuralClassifier:
|
30 |
+
|
31 |
+
def __init__(self, model_path: str, uses_context: bool, device: str):
|
32 |
+
self.config = AutoConfig.from_pretrained(model_path)
|
33 |
+
self.device = device
|
34 |
+
self.model = AutoModelForSequenceClassification.from_pretrained(model_path, config=self.config).to(device)
|
35 |
+
self.tokenizer = AutoTokenizer.from_pretrained(model_path)
|
36 |
+
self.uses_context = uses_context
|
37 |
+
|
38 |
+
def predict_sentence(self, sentence: str, context: str = None):
|
39 |
+
if context is None and self.uses_context:
|
40 |
+
raise ValueError("You need to pass in context argument, including the sentence")
|
41 |
+
|
42 |
+
features = self.tokenizer(sentence, text_pair=context,
|
43 |
+
padding="max_length", truncation=True, return_tensors='pt')
|
44 |
+
outputs = self.model(**features.to(self.device), return_dict=True)
|
45 |
+
argmax = outputs.logits.argmax(dim=-1).detach().cpu().tolist()[0]
|
46 |
+
labels = LABELS[argmax]
|
47 |
+
|
48 |
+
return labels
|
49 |
+
```
|
50 |
+
|
51 |
+
The wrapper can be used as follows:
|
52 |
+
```python
|
53 |
+
classifier = NeuralClassifier(model_path="MU-NLPC/XLM-R-large-reflective-conf4",
|
54 |
+
uses_context=False,
|
55 |
+
device="cpu")
|
56 |
+
|
57 |
+
test_sentences = ["And one day I will be a real teacher and I will try to do the best I can for the children.",
|
58 |
+
"I felt really well!",
|
59 |
+
"gfagdhj gjfdjgh dg"]
|
60 |
+
|
61 |
+
y_pred = [classifier.predict_sentence(sentence) for sentence in tqdm(test_sentences)]
|
62 |
+
|
63 |
+
print(y_pred)
|
64 |
+
|
65 |
+
>>> ['Intention', 'Feeling', 'Other']
|
66 |
+
```
|
67 |
+
|
68 |
+
### Training Data
|
69 |
+
|
70 |
+
The model was trained on a [CEReD dataset](http://hdl.handle.net/11372/LRT-3573) and aims for the best possible evaluation in cross-lingual settings (on unseen languages).
|
71 |
+
|
72 |
+
See the reproducible training script in the project directory: https://github.com/EduMUNI/reflection-classification
|
73 |
+
|
74 |
+
## Citation
|
75 |
+
|
76 |
+
If you use the model in scientific work, please acknowledge our work as follows.
|
77 |
+
|
78 |
+
```bibtex
|
79 |
+
@Article{Nehyba2022applications,
|
80 |
+
author={Nehyba, Jan and {\v{S}}tef{\'a}nik, Michal},
|
81 |
+
title={Applications of deep language models for reflective writings},
|
82 |
+
journal={Education and Information Technologies},
|
83 |
+
year={2022},
|
84 |
+
month={Sep},
|
85 |
+
day={05},
|
86 |
+
issn={1573-7608},
|
87 |
+
doi={10.1007/s10639-022-11254-7},
|
88 |
+
url={https://doi.org/10.1007/s10639-022-11254-7}
|
89 |
+
}
|
90 |
+
```
|