search

thunlp / SememePSO-Attack

Code and data of the ACL 2020 paper "Word-level Textual Adversarial Attacking as Combinatorial Optimization"
MIT License
86 stars 14 forks source link

delemmatization #6

Closed dp0d closed 2 months ago

dp0d commented 3 years ago

Hi, thank you for provide the code! May I ask you that how you guys do the delemmatization in the code of IMDB dataset?

zangy17 commented 3 years ago

We build a dictionary of the prototype words and their inflections occurred in IMDB dataset. Then we can exert the same inflections on the substitutions.

dp0d commented 3 years ago

oo……, In the code, I only found the operation that using prototype words to find their candidates rather than doing the delematization. Could you please inform me the exact steps? I'd appreaciate it if you can update the code. image

wenkfjsf commented 2 months ago

有人成功复现了吗

yangalan123 commented 2 months ago

Hi all,

Thanks for all your follow-ups on this issue and sorry for the late responses as all main authors have graduated and have not worked in this direction recently. We can guarantee that our experiments are replicable ( “可复现的" in Mandarin). There are already community-built toolkits that have successfully replicated our experiments, and improved our code quality and running efficiency by a lot: TextAttack (EMNLP 2020 Demo, ~3k stars), and OpenAttack (ACL 2021 Demo, ~600 stars). So we encourage you to check the latest improved version there. This repo is currently under archival mode so the response would be significantly delayed and no major maintenance will happen in the foreseeable future. Thanks for your interest in this work and we encourage you to try the aforementioned toolkits and ask questions there.

Best, Chenghao