Enhanced transformer for length-controlled abstractive summarization based on summary output area

Area of Science:

• Natural Language Processing
• Computer Vision
• Artificial Intelligence

Background:

• Current abstractive summarization models generate single summaries, lacking precise length control.
• Existing length-controllable methods struggle to select relevant information within length constraints.

Purpose of the Study:

• To develop a novel approach for length-controllable abstractive summarization.
• To integrate an image processing phase for determining summary output slot size.
• To adapt enhanced T5 and GPT models for generating summaries that perfectly fit designated slots.

Main Methods:

• A novel approach integrating an image processing phase to determine summary output slot size.
• Utilizing enhanced T5 and GPT models for abstractive summarization.
• Employing the computed area of a slot to tailor summary generation.

Main Results:

• The proposed model successfully generates abstractive summaries tailored to fit specific output slots.
• Experimental evaluations on the CNN/Daily Mail dataset demonstrate superior length-controlled summarization performance.
• The image processing integration effectively guides summary length adaptation.

Conclusions:

• The novel image processing-integrated approach offers effective length control for abstractive summarization.
• Enhanced T5 and GPT models can be adapted to produce contextually relevant, length-specific summaries.
• This method advances practical applications requiring precise summary length adaptation.