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")

input = ["Ale fajnie, spadł dzisiaj śnieg! Ulepimy dziś bałwana?"]

encoding = tokenizer(
          input,
          add_special_tokens=True,
          return_token_type_ids=True,
          truncation=True,
          padding='max_length',
          return_attention_mask=True,
          return_tensors='pt',
        )
output = model(**encoding).logits.to("cpu").detach().numpy()
prediction = id2label[np.argmax(output)]
print(input, "--->", prediction)

Predicted output:

['Ale fajnie, spadł dzisiaj śnieg! Ulepimy dziś bałwana?'] ---> positive

Overview

• Language model: allegro/herbert-base-cased
• Language: pl
• Training data: Reviews + own data
• Blog post: Sentiment analysis - COVID-19 – the source of the heated discussion