- What are approaches for summarizing large texts?
Current models such as PEGASUS or GPT-2 have the limit for the number of input tokens, meaning large texts cannot be summarized in one pass. What are approaches for large texts?
Wu et al., “Recursively Summarizing Books with Human Feedback”, (2021).
- Why PEGASUS is so slow?
It takes over a minute to summarize 5 relatively short texts. "PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization", 2020
- How to fine tune summaries for small data sets?
E.g. CULPA reviews that describe professors not as bulletpoints of newspapers, but by generating descriptions of what a person like. Keywords: few shots learning.
Passali "Towards Human-Centered Summarization: A Case Study on Financial News", 2021.