search

csebuetnlp / xl-sum

This repository contains the code, data, and models of the paper titled "XL-Sum: Large-Scale Multilingual Abstractive Summarization for 44 Languages" published in Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021.
https://aclanthology.org/2021.findings-acl.413/
257 stars 41 forks source link

when the text is very short #4

Closed 08tjlys closed 3 years ago

08tjlys commented 3 years ago

hi, first of all, really thanks for such great project. When i tried your inference code on hugging face, i found that when the text is short, for example like just "weather is fine", it will generate a longer unrelated summary. i am totally new to summarization, so is there any thing i should pay attention to in such case? for example, in such case, generating inputs_ids will be different? really appreciate your guide if possible, thanks!

Tahmid04 commented 3 years ago

Hi, thanks for appreciating our work. Since XL-Sum is a news article-summary dataset, models pretrained on it are expected to generate news-like summaries, even when the input is not from the news domain, e.g., paper abstracts, and for your case non-article inputs. If you want to generate summaries for non-news inputs, we recommend you further fine-tune the model checkpoint on such data. And if you just want to reduce the output length, using a higher length penalty may suit your use case. Hope that helps!

08tjlys commented 3 years ago

thanks a lot, it really helps