|
--- |
|
pipeline_tag: text-classification |
|
widget: |
|
- text: "Pani Katarzyno z jakiej racji moja paczka przyszła do sąsiada zamiast do mnie? Nie można poprawnie nadać paczki?" |
|
example_title: "Sentiment" |
|
license: cc-by-4.0 |
|
language: |
|
- pl |
|
--- |
|
|
|
<img src="https://public.3.basecamp.com/p/rs5XqmAuF1iEuW6U7nMHcZeY/upload/download/VL-NLP-short.png" alt="logo voicelab nlp" style="width:300px;"/> |
|
|
|
# Sentiment Classification in Polish |
|
|
|
``` |
|
import numpy as np |
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
|
id2label = {0: "negative", 1: "neutral", 2: "positive"} |
|
tokenizer = AutoTokenizer.from_pretrained("Voicelab/herbert-base-cased-sentiment") |
|
|
|
model = AutoModelForSequenceClassification.from_pretrained("Voicelab/herbert-base-cased-sentiment") |
|
|
|
encoding = tokenizer( |
|
input, |
|
add_special_tokens=True, |
|
return_token_type_ids=True, |
|
truncation=True, |
|
padding='max_length', |
|
return_attention_mask=True, |
|
return_tensors='pt', |
|
).to("cuda:0") |
|
output = model(**encoding).logits.to("cpu").detach().numpy() |
|
prediction = id2label[np.argmax(output)] |
|
|
|
``` |
|
|
|
|
|
### Overview |
|
- **Language model:** [allegro/herbert-base-cased](https://huggingface.co/allegro/herbert-base-cased) |
|
- **Language:** pl |
|
- **Training data:** Reviews + own data |
|
- **Blog post:** [Sentiment analysis - COVID-19 – the source of the heated discussion](https://voicelab.ai/covid-19-the-source-of-the-heated-discussion) |
|
|
|
|
