deberta v3 base absa v1.1

Introduction

The deberta-v3-base-absa-v1.1 model is designed for aspect-based sentiment analysis using the DeBERTa architecture. It is implemented in PyTorch and utilizes Safetensors for efficient data handling. The model is compatible with several datasets and is primarily focused on English language text classification tasks.

Architecture

This model builds upon the DeBERTa V3 architecture, specifically leveraging the microsoft/deberta-v3-base variant. It is enhanced with the FAST-LCF-BERT model from the PyABSA framework, which is an open-source tool for performing aspect-based sentiment analysis.

Training

The model is fine-tuned on a substantial dataset comprising over 180k examples, including data augmentation techniques. The training datasets include Laptop14, Restaurant14, Restaurant16, ACL Twitter, MAMS, Television, TShirt, and Yelp, among others. The model was trained to improve accuracy and macro F1 metrics.

Guide: Running Locally

1. Installation:

   • Use the Hugging Face Transformers library to load the model.

   from transformers import AutoTokenizer, AutoModelForSequenceClassification
   
   model_name = "yangheng/deberta-v3-base-absa-v1.1"
   tokenizer = AutoTokenizer.from_pretrained(model_name)
   model = AutoModelForSequenceClassification.from_pretrained(model_name)
   

2. Usage:

   • Implement the text classification pipeline.

   from transformers import pipeline
   
   classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)
   
   for aspect in ['camera', 'phone']:
      print(aspect, classifier('The camera quality of this phone is amazing.', text_pair=aspect))
   

3. Hardware:

   • For optimal performance, especially during training or handling large datasets, consider leveraging cloud GPUs such as those available on AWS, Google Cloud, or Azure.

License

The deberta-v3-base-absa-v1.1 model is licensed under the MIT License, allowing for wide usage and modification within the constraints of this permissive license.