Three major corpus tools mis-tokenize emojis and homoglyphs, and a proposed preprocessing pipeline (retokenization, emoji transliteration, NFKC normalization) eliminates the resulting mismatches on tested data.
running word
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Data interference: emojis, homoglyphs, and issues of data fidelity in corpora and their results
Three major corpus tools mis-tokenize emojis and homoglyphs, and a proposed preprocessing pipeline (retokenization, emoji transliteration, NFKC normalization) eliminates the resulting mismatches on tested data.