Word Level Noising - Githubissues

djaszak / NLPAug

A framework to simplify the usage of common Data Augmentation methods in the NLP context.

MIT License

2 stars 0 forks source link

Word Level Noising #8

Open djaszak opened 2 years ago

djaszak commented 2 years ago

[ ] Unigram noising -> Replacing words by different random words
[ ] Blank noising -> Replacing words by "_"
[ ] Syntactic noise -> Shortening, alteration of adjectives
[ ] Semantic noise -> Lexical substitution of synonyms (See next point)
[ ] Random swap (EDA)
[ ] Random deletion (EDA)
[ ] Noise instead of zero-padding
[ ] TF-IDF -> Replace uninformative words by other uninformative

djaszak commented 2 years ago

With “unigram noising”, words in the input data are replaced by another word with a certain probability. By
the method of “blank noising”, words get replaced with “_”. By the adoption of both patterns, the authors achieved
improved results in their experiments.

From "Markus Bayer, Marc-André Kaufhold, Christian Reuter (2021) A Survey on Data Augmentation for Text Classification"