Dataset-Aware Automated Model Selection for Abstractive Summarization: A Meta-Learning Approach

DOI

10.2991/978-94-6239-713-2_43How to use a DOI?

Keywords

Automated Model Selection; Meta Learning; Abstractive Summarization; AutoML

Abstract

Pretrained transformer-based models have significantly advanced abstractive summarization. However, selecting the right model for a new dataset remains challenging because exhaustively fine-tuning multiple candidate models for each dataset is computationally expensive and limits the scalability of summarization pipelines. This paper presents a dataset-aware meta learning framework for automated model selection in this context. We first construct an empirical oracle by fine-tuning key transformer-based models under a standardized training budget across heterogeneous summarization datasets. Model performance is evaluated using both lexical (ROUGE-L) and semantic (SBERT cosine similarity) metrics, producing objective-aware oracle labels. Dataset-level meta-features are then used to train supervised selectors that predict suitable models for previously unseen datasets using the LODO strategy. Experimental results reveal substantial dataset-dependent variability and a 71.4% disagreement between lexical and semantic oracle models, highlighting the need for automatic model selection. The proposed meta-learning selector achieves up to 85.7% accuracy in predicting lexical oracle models and 71% accuracy for semantic objectives, while significantly reducing the need for exhaustive fine-tuning.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Dhivya Bino
AU  - Manish Shrivastava
PY  - 2026
DA  - 2026/06/25
TI  - Dataset-Aware Automated Model Selection for Abstractive Summarization: A Meta-Learning Approach
BT  - Proceedings of the International Conference on Advances in Computing Technology and Artificial Intelligence (COMPUTATIA 2026)
PB  - Atlantis Press
SP  - 578
EP  - 590
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-713-2_43
DO  - 10.2991/978-94-6239-713-2_43
ID  - Bino2026
ER  -