README.md
| 1 | --- |
| 2 | language: en |
| 3 | widget: |
| 4 | - text: Covid cases are increasing fast! |
| 5 | datasets: |
| 6 | - tweet_eval |
| 7 | license: cc-by-4.0 |
| 8 | --- |
| 9 | |
| 10 | |
| 11 | # Twitter-roBERTa-base for Sentiment Analysis - UPDATED (2022) |
| 12 | |
| 13 | 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. |
| 14 | The original Twitter-based RoBERTa model can be found [here](https://huggingface.co/cardiffnlp/twitter-roberta-base-2021-124m) and the original reference paper is [TweetEval](https://github.com/cardiffnlp/tweeteval). This model is suitable for English. |
| 15 | |
| 16 | - Reference Paper: [TimeLMs paper](https://arxiv.org/abs/2202.03829). |
| 17 | - Git Repo: [TimeLMs official repository](https://github.com/cardiffnlp/timelms). |
| 18 | |
| 19 | <b>Labels</b>: |
| 20 | 0 -> Negative; |
| 21 | 1 -> Neutral; |
| 22 | 2 -> Positive |
| 23 | |
| 24 | This sentiment analysis model has been integrated into [TweetNLP](https://github.com/cardiffnlp/tweetnlp). You can access the demo [here](https://tweetnlp.org). |
| 25 | |
| 26 | ## Example Pipeline |
| 27 | ```python |
| 28 | from transformers import pipeline |
| 29 | sentiment_task = pipeline("sentiment-analysis", model=model_path, tokenizer=model_path) |
| 30 | sentiment_task("Covid cases are increasing fast!") |
| 31 | ``` |
| 32 | ``` |
| 33 | [{'label': 'Negative', 'score': 0.7236}] |
| 34 | ``` |
| 35 | |
| 36 | ## Full classification example |
| 37 | |
| 38 | ```python |
| 39 | from transformers import AutoModelForSequenceClassification |
| 40 | from transformers import TFAutoModelForSequenceClassification |
| 41 | from transformers import AutoTokenizer, AutoConfig |
| 42 | import numpy as np |
| 43 | from scipy.special import softmax |
| 44 | # Preprocess text (username and link placeholders) |
| 45 | def preprocess(text): |
| 46 | new_text = [] |
| 47 | for t in text.split(" "): |
| 48 | t = '@user' if t.startswith('@') and len(t) > 1 else t |
| 49 | t = 'http' if t.startswith('http') else t |
| 50 | new_text.append(t) |
| 51 | return " ".join(new_text) |
| 52 | MODEL = f"cardiffnlp/twitter-roberta-base-sentiment-latest" |
| 53 | tokenizer = AutoTokenizer.from_pretrained(MODEL) |
| 54 | config = AutoConfig.from_pretrained(MODEL) |
| 55 | # PT |
| 56 | model = AutoModelForSequenceClassification.from_pretrained(MODEL) |
| 57 | #model.save_pretrained(MODEL) |
| 58 | text = "Covid cases are increasing fast!" |
| 59 | text = preprocess(text) |
| 60 | encoded_input = tokenizer(text, return_tensors='pt') |
| 61 | output = model(**encoded_input) |
| 62 | scores = output[0][0].detach().numpy() |
| 63 | scores = softmax(scores) |
| 64 | # # TF |
| 65 | # model = TFAutoModelForSequenceClassification.from_pretrained(MODEL) |
| 66 | # model.save_pretrained(MODEL) |
| 67 | # text = "Covid cases are increasing fast!" |
| 68 | # encoded_input = tokenizer(text, return_tensors='tf') |
| 69 | # output = model(encoded_input) |
| 70 | # scores = output[0][0].numpy() |
| 71 | # scores = softmax(scores) |
| 72 | # Print labels and scores |
| 73 | ranking = np.argsort(scores) |
| 74 | ranking = ranking[::-1] |
| 75 | for i in range(scores.shape[0]): |
| 76 | l = config.id2label[ranking[i]] |
| 77 | s = scores[ranking[i]] |
| 78 | print(f"{i+1}) {l} {np.round(float(s), 4)}") |
| 79 | ``` |
| 80 | |
| 81 | Output: |
| 82 | |
| 83 | ``` |
| 84 | 1) Negative 0.7236 |
| 85 | 2) Neutral 0.2287 |
| 86 | 3) Positive 0.0477 |
| 87 | ``` |
| 88 | |
| 89 | |
| 90 | ### References |
| 91 | ``` |
| 92 | @inproceedings{camacho-collados-etal-2022-tweetnlp, |
| 93 | title = "{T}weet{NLP}: Cutting-Edge Natural Language Processing for Social Media", |
| 94 | author = "Camacho-collados, Jose and |
| 95 | Rezaee, Kiamehr and |
| 96 | Riahi, Talayeh and |
| 97 | Ushio, Asahi and |
| 98 | Loureiro, Daniel and |
| 99 | Antypas, Dimosthenis and |
| 100 | Boisson, Joanne and |
| 101 | Espinosa Anke, Luis and |
| 102 | Liu, Fangyu and |
| 103 | Mart{\'\i}nez C{\'a}mara, Eugenio" and others, |
| 104 | booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing: System Demonstrations", |
| 105 | month = dec, |
| 106 | year = "2022", |
| 107 | address = "Abu Dhabi, UAE", |
| 108 | publisher = "Association for Computational Linguistics", |
| 109 | url = "https://aclanthology.org/2022.emnlp-demos.5", |
| 110 | pages = "38--49" |
| 111 | } |
| 112 | |
| 113 | ``` |
| 114 | |
| 115 | ``` |
| 116 | @inproceedings{loureiro-etal-2022-timelms, |
| 117 | title = "{T}ime{LM}s: Diachronic Language Models from {T}witter", |
| 118 | author = "Loureiro, Daniel and |
| 119 | Barbieri, Francesco and |
| 120 | Neves, Leonardo and |
| 121 | Espinosa Anke, Luis and |
| 122 | Camacho-collados, Jose", |
| 123 | booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics: System Demonstrations", |
| 124 | month = may, |
| 125 | year = "2022", |
| 126 | address = "Dublin, Ireland", |
| 127 | publisher = "Association for Computational Linguistics", |
| 128 | url = "https://aclanthology.org/2022.acl-demo.25", |
| 129 | doi = "10.18653/v1/2022.acl-demo.25", |
| 130 | pages = "251--260" |
| 131 | } |
| 132 | |
| 133 | ``` |