cardiffnlp/twitter-roberta-base-topic-latest

This is a RoBERTa-base model trained on 154M tweets until the end of December 2022 and finetuned for topic classification (multilabel classification) on the TweetTopic dataset of SuperTweetEval. The original Twitter-based RoBERTa model can be found here.

Labels

"id2label": { "0": "arts_&_culture", "1": "business_&_entrepreneurs", "2": "celebrity_&_pop_culture", "3": "diaries_&_daily_life", "4": "family", "5": "fashion_&_style", "6": "film_tv_&_video", "7": "fitness_&_health", "8": "food_&_dining", "9": "gaming", "10": "learning_&_educational", "11": "music", "12": "news_&_social_concern", "13": "other_hobbies", "14": "relationships", "15": "science_&_technology", "16": "sports", "17": "travel_&_adventure", "18": "youth_&_student_life" }

Example

from transformers import pipeline
text = "So @AB is just the latest victim of the madden curse. If you’re on the cover of that game your career will take a turn for the worse"

pipe = pipeline('text-classification', model="cardiffnlp/twitter-roberta-base-topic-latest", return_all_scores=True)
predictions = pipe(text)[0]
predictions = [x for x in predictions if x['score'] > 0.5]
predictions
>> [{'label': 'gaming', 'score': 0.899931013584137},
 {'label': 'sports', 'score': 0.5215537548065186}]

Citation Information

Please cite the reference paper if you use this model.

@inproceedings{antypas2023supertweeteval,
  title={SuperTweetEval: A Challenging, Unified and Heterogeneous Benchmark for Social Media NLP Research},
  author={Dimosthenis Antypas and Asahi Ushio and Francesco Barbieri and Leonardo Neves and Kiamehr Rezaee and Luis Espinosa-Anke and Jiaxin Pei and Jose Camacho-Collados},
  booktitle={Findings of the Association for Computational Linguistics: EMNLP 2023},
  year={2023}
}