Automated Categorization of Pre-trained Models for Software Engineering

This paper is a research article that presents a semi-automated approach for categorizing pre-trained models (PTMs) specifically for software engineering (SE) tasks using a dataset from the Hugging Face platform. The study addresses the challenge faced by users with limited expertise in selecting appropriate PTMs for their tasks, given that existing categorizations on the platform are more aligned with generic machine learning categories. The authors outline a methodology that involves extracting PTM information from Hugging Face, identifying SE tasks from the literature, and creating a mapping between HF tags and specific SE tasks. The evaluation indicates that model cards provide sufficient information for classifying PTMs based on pipeline tags. Furthermore, the paper discusses the importance of developing advanced techniques for recommending suitable PTMs for specific SE tasks, which could lead to the creation of dedicated recommender systems for software engineering. The findings aim to enhance the accessibility and usability of PTMs for developers engaged in software engineering activities.