cardiffnlp/roberta-large-tweet-topic-multi-2020

这个模型是在 tweet_topic_multi 上进行微调的 roberta-large 的版本。该模型在 train_2020 数据集上进行微调，并在 test_2021 数据集上进行验证。微调脚本可以在 here 找到。该模型在 test_2021 数据集上取得以下结果:

• F1 (micro): 0.7323655694132079
• F1 (macro): 0.5794562917377284
• 准确度: 0.4937462775461584

用法

import math
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

def sigmoid(x):
  return 1 / (1 + math.exp(-x))
  
tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/roberta-large-tweet-topic-multi-2020")
model = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/roberta-large-tweet-topic-multi-2020", problem_type="multi_label_classification")
model.eval()
class_mapping = model.config.id2label

with torch.no_grad():
  text = #NewVideo Cray Dollas- Water- Ft. Charlie Rose- (Official Music Video)- {{URL}} via {@YouTube@} #watchandlearn {{USERNAME}}
  tokens = tokenizer(text, return_tensors='pt')
  output = model(**tokens)
  flags = [sigmoid(s) > 0.5 for s in output[0][0].detach().tolist()]
  topic = [class_mapping[n] for n, i in enumerate(flags) if i]
print(topic)

参考资料

@inproceedings{dimosthenis-etal-2022-twitter,
    title = "{T}witter {T}opic {C}lassification",
    author = "Antypas, Dimosthenis  and
    Ushio, Asahi  and
    Camacho-Collados, Jose  and
    Neves, Leonardo  and
    Silva, Vitor  and
    Barbieri, Francesco",
    booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "International Committee on Computational Linguistics"
}