T5-base fine-tuned for Sentiment Span Extraction

Google's T5 base fine-tuned on Tweet Sentiment Extraction Dataset for Span Sentiment Extraction downstream task.

Details of T5

The T5 model was presented in Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer by Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu in Here the abstract:

Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts every language problem into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled datasets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new “Colossal Clean Crawled Corpus”, we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our dataset, pre-trained models, and code.

Details of the downstream task (Span Sentiment Extraction) - Dataset 📚

Tweet Sentiment Extraction Dataset

"My ridiculous dog is amazing." [sentiment: positive]

With all of the tweets circulating every second it is hard to tell whether the sentiment behind a specific tweet will impact a company, or a person's, brand for being viral (positive), or devastate profit because it strikes a negative tone. Capturing sentiment in language is important in these times where decisions and reactions are created and updated in seconds. But, which words actually lead to the sentiment description? In this competition you will need to pick out the part of the tweet (word or phrase) that reflects the sentiment.

Help build your skills in this important area with this broad dataset of tweets. Work on your technique to grab a top spot in this competition. What words in tweets support a positive, negative, or neutral sentiment? How can you help make that determination using machine learning tools?

In this competition we've extracted support phrases from Figure Eight's Data for Everyone platform. The dataset is titled Sentiment Analysis: Emotion in Text tweets with existing sentiment labels, used here under creative commons attribution 4.0. international licence. Your objective in this competition is to construct a model that can do the same - look at the labeled sentiment for a given tweet and figure out what word or phrase best supports it.

Disclaimer: The dataset for this competition contains text that may be considered profane, vulgar, or offensive.

Dataset	Split	# samples
TSE	train	23907
TSE	eval	3573

Model fine-tuning 🏋️‍

The training script is a slightly modified version of this Colab Notebook created by Lorenzo Ampil, so all credits to him!

Model in Action 🚀

from transformers import AutoModelWithLMHead, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-span-sentiment-extraction")
model = AutoModelWithLMHead.from_pretrained("mrm8488/t5-base-finetuned-span-sentiment-extraction")

def get_sentiment_span(text):
  input_ids = tokenizer.encode(text, return_tensors="pt", add_special_tokens=True)  # Batch size 1
  
  generated_ids = model.generate(input_ids=input_ids, num_beams=1, max_length=80).squeeze()
  
  predicted_span = tokenizer.decode(generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
    
  return predicted_span
  
get_sentiment_span("question: negative context: My bike was put on hold...should have known that.... argh total bummer")

# output: 'argh total bummer'

get_sentiment_span("question: positive context: On the monday, so i wont be able to be with you! i love you")

# output: 'i love you'

Created by Manuel Romero/@mrm8488 | LinkedIn

Made with ♥ in Spain