S

Sciroshot

Developed by BSC-LT
Zero-shot text classification model based on RoBERTa-large, optimized for scientific domains with excellent performance in cross-domain tasks
Downloads 83
Release Time : 2/8/2023

Model Overview

Zero-shot classification model fine-tuned using Microsoft Academic Graph (MAG) scientific paper data, employing entailment methods for efficient text classification

Model Features

Scientific domain optimization
Fine-tuned using Microsoft Academic Graph scientific paper data, achieving state-of-the-art performance in scientific classification tasks
Cross-domain generalization
While optimized for scientific domains, it still performs excellently in general domain classification tasks
Entailment approach
Innovatively transforms classification tasks into entailment relationship judgments between text and labels
Efficient training strategy
Prevents overfitting through early stopping, significantly improving zero-shot classification performance

Model Capabilities

Zero-shot text classification
Multi-label classification
Scientific literature classification
Cross-domain text classification

Use Cases

Academic research
Scientific paper classification
Automatically classify research papers into 292 scientific domain categories
Achieved 42.22% accuracy on arXiv dataset
Literature retrieval enhancement
Improve literature classification and recommendation functions for academic search engines
Content management
News classification
Automatic multi-topic classification of news articles
Achieved 59.08% accuracy on Yahoo Answers topic classification
Social media content analysis
Identify topic categories of social media posts
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase