Twitter-roBERTa-base for Sentiment Analysis – UPDATED (2022)

This is a RoBERTa-base model trained on ~124M tweets from January 2018 to December 2021, and finetuned for sentiment analysis with the TweetEval benchmark.
The original Twitter-based RoBERTa model can be found here and the original reference paper is TweetEval. This model is suitable for English.

• Reference Paper: TimeLMs paper.
• Git Repo: TimeLMs official repository.

Labels:
0 -> Negative;
1 -> Neutral;
2 -> Positive
This sentiment analysis model has been integrated into TweetNLP. You can access the demo here.

Example Pipeline

from transformers import pipeline<br /> sentiment_task = pipeline("sentiment-analysis", model=model_path, tokenizer=model_path)<br /> sentiment_task("Covid cases are increasing fast!")<br />

[{'label': 'Negative', 'score': 0.7236}]<br />

Full classification example

from transformers import AutoModelForSequenceClassification<br /> from transformers import TFAutoModelForSequenceClassification<br /> from transformers import AutoTokenizer, AutoConfig<br /> import numpy as np<br /> from scipy.special import softmax<br /> # Preprocess text (username and link placeholders)<br /> def preprocess(text):<br /> new_text = []<br /> for t in text.split(" "):<br /> t = '@user' if t.startswith('@') and len(t) > 1 else t<br /> t = 'http' if t.startswith('http') else t<br /> new_text.append(t)<br /> return " ".join(new_text)<br /> MODEL = f"cardiffnlp/twitter-roberta-base-sentiment-latest"<br /> tokenizer = AutoTokenizer.from_pretrained(MODEL)<br /> config = AutoConfig.from_pretrained(MODEL)<br /> # PT<br /> model = AutoModelForSequenceClassification.from_pretrained(MODEL)<br /> #model.save_pretrained(MODEL)<br /> text = "Covid cases are increasing fast!"<br /> text = preprocess(text)<br /> encoded_input = tokenizer(text, return_tensors='pt')<br /> output = model(**encoded_input)<br /> scores = output[0][0].detach().numpy()<br /> scores = softmax(scores)<br /> # # TF<br /> # model = TFAutoModelForSequenceClassification.from_pretrained(MODEL)<br /> # model.save_pretrained(MODEL)<br /> # text = "Covid cases are increasing fast!"<br /> # encoded_input = tokenizer(text, return_tensors='tf')<br /> # output = model(encoded_input)<br /> # scores = output[0][0].numpy()<br /> # scores = softmax(scores)<br /> # Print labels and scores<br /> ranking = np.argsort(scores)<br /> ranking = ranking[::-1]<br /> for i in range(scores.shape[0]):<br /> l = config.id2label[ranking[i]]<br /> s = scores[ranking[i]]<br /> print(f"{i+1}) {l} {np.round(float(s), 4)}")<br />

Output:
1) Negative 0.7236<br /> 2) Neutral 0.2287<br /> 3) Positive 0.0477<br />