metadata

language:
  - en
tags:
  - text-classification
  - pytorch
metrics:
  - accuracy
  - f1-score
extra_gated_prompt: >-
  Our models are intended for academic use only. If you are not affiliated with
  an academic institution, please provide a rationale for using our models.

  If you use our models for your work or research, please cite this paper:
  Sebők, M., Máté, Á., Ring, O., Kovács, V., & Lehoczki, R. (2024). Leveraging
  Open Large Language Models for Multilingual Policy Topic Classification: The
  Babel Machine Approach. Social Science Computer Review, 0(0).
  https://doi.org/10.1177/08944393241259434
extra_gated_fields:
  Name: text
  Country: country
  Institution: text
  E-mail: text
  Use case: text

xlm-roberta-large-english-legislative-cap-v3

Model description

An xlm-roberta-large model fine-tuned on english training data containing legislative documents (bills, laws, motions, legislative decrees, hearings, resolutions) labeled with major topic codes from the Comparative Agendas Project.

How to use the model

This snippet prints the three most probable labels and their corresponding softmax scores:

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("poltextlab/xlm-roberta-large-english-legislative-cap-v3")
tokenizer = AutoTokenizer.from_pretrained("xlm-roberta-large")

sentence = "This is an example."

inputs = tokenizer(sentence,
                   return_tensors="pt",
                   max_length=512,
                   padding="do_not_pad",
                   truncation=True
                   )

logits = model(**inputs).logits

probs = torch.softmax(logits, dim=1).tolist()[0]
probs = {model.config.id2label[index]: round(probability, 2) for index, probability in enumerate(probs)}
top3_probs = dict(sorted(probs.items(), key=lambda item: item[1], reverse=True)[:3])

print(top3_probs)

Model performance

The model was evaluated on a test set of 148474 examples.
Model accuracy is 0.9.

label	precision	recall	f1-score	support
0	0.85	0.8	0.83	5794
1	0.83	0.81	0.82	2920
2	0.91	0.92	0.92	10336
3	0.9	0.92	0.91	5258
4	0.82	0.9	0.86	4604
5	0.91	0.93	0.92	6173
6	0.87	0.89	0.88	5111
7	0.88	0.92	0.9	4251
8	0.87	0.91	0.89	1517
9	0.89	0.92	0.91	8119
10	0.91	0.9	0.9	10326
11	0.87	0.88	0.88	5471
12	0.86	0.86	0.86	3078
13	0.88	0.86	0.87	9050
14	0.88	0.89	0.88	10197
15	0.87	0.88	0.87	2556
16	0.92	0.92	0.92	4821
17	0.86	0.86	0.86	4106
18	0.9	0.87	0.89	17295
19	0.9	0.9	0.9	10681
20	1	0.2	0.33	25
21	0.99	0.96	0.97	16785
macro avg	0.89	0.86	0.86	148474
weighted avg	0.9	0.9	0.9	148474

Fine-tuning procedure

This model was fine-tuned with the following key hyperparameters:

Number of Training Epochs: 10
Batch Size: 8
Learning Rate: 5e-06
Early Stopping: enabled with a patience of 2 epochs

Inference platform

This model is used by the CAP Babel Machine, an open-source and free natural language processing tool, designed to simplify and speed up projects for comparative research.

Cooperation

Model performance can be significantly improved by extending our training sets. We appreciate every submission of CAP-coded corpora (of any domain and language) at poltextlab{at}poltextlab{dot}com or by using the CAP Babel Machine.

Reference

Sebők, M., Máté, Á., Ring, O., Kovács, V., & Lehoczki, R. (2024). Leveraging Open Large Language Models for Multilingual Policy Topic Classification: The Babel Machine Approach. Social Science Computer Review, 0(0). https://doi.org/10.1177/08944393241259434

Debugging and issues

This architecture uses the sentencepiece tokenizer. In order to use the model before transformers==4.27 you need to install it manually.

If you encounter a RuntimeError when loading the model using the from_pretrained() method, adding ignore_mismatched_sizes=True should solve the issue.