Summarization · TOPSIS
Choosing a summarizer through multiple criteria.
A comparative study using TOPSIS to rank four pretrained summarization models across quality and efficiency-related criteria.
The problem
Choosing a summarization model from one metric ignores the tradeoffs between different kinds of overlap, loss, and output length.
How it comes together
A decision matrix compares BART, IT5, Pegasus, and LED. Weighted criteria are normalized, then distance from ideal best and worst alternatives determines the ranking.
Follow the flow.
Four pretrained transformer models.
ROUGE-1, ROUGE-2, ROUGE-L, loss, and generation-length criteria.
Explicit criterion weights and benefit/cost directions.
An inspectable TOPSIS score and ranked results.
Why this approach?
Make the weighting assumptions explicit. The resulting ranking reflects this study's chosen criteria and data, rather than claiming a universal best model.