One row per emoji, with the cues that tell what it is doing

emoji_features() returns one row per emoji occurrence with the cues a classifier of emoji function reads: where the emoji sits, what is on either side of it, the run it belongs to, and what kind of emoji it is. It reports cues and withholds labels. The best published classifier of what an emoji does in a sentence reached an F1 of 0.62 on content and its annotators a Fleiss kappa of 0.50 on subtype (Na'aman, Provenza and Montoya 2017), and a shipped label at that accuracy is worse than none. The Unicode group, the cue that helped that classifier most, is a column.

Usage

emoji_features(data, text, presentation = "emoji")

Arguments

data

A data frame or tibble containing a text column. Grouped data frames are accepted. The verbs that work a row at a time (adding columns, or keeping and expanding rows) carry the grouping through to their result, as dplyr::mutate() and dplyr::filter() do. The verbs that pool across rows – the counts, the co-occurrence edge lists, the time series – warn that they ignore the grouping and return one corpus-wide answer.

text

The text column to scan, supplied unquoted. Any atomic column is accepted and read as character, so a factor works and a numeric, Date or logical one simply contains no emoji. A list column – or a data-frame column – is refused rather than coerced, because coercing one deparses it and the emoji found would be in the code rather than in your data. What counts as an emoji is the same in every verb; see the Detection section of tidyEmoji for the one case that surprises people, code points that are emoji only when they carry U+FE0F.

presentation

Which code points count as emoji: "emoji" (the default) or "any".

• "emoji" requires emoji presentation, so U+2764 U+FE0F is an emoji and the bare U+2764 is not. This is the default and it will stay the default: widening it would silently recount every corpus already analysed.

• "any" also reads the text-presentation spellings Unicode lists for an emoji (the bare heart, U+263A, U+2714, a keycap without its selector), because several keyboards emit them. It never reads the copyright, registered and trade-mark signs (U+00A9, U+00AE, U+2122), which are typography far more often than they are emoji.

"any" only ever adds occurrences: everything "emoji" finds, it finds too, in the same place. emoji_coverage() reports how many it adds and how many of those a lexicon can score, which is the cost of the choice in one number. Every verb that counts, scores or rewrites the emoji in a text column takes this argument, so counts still agree across the package for whichever value you use. The flag verbs do not need it, since a flag has no text-presentation form.

Value

A tibble with one row per emoji occurrence, carrying the columns of data, and .row_number, .emoji_index (its number within the row), .emoji, .emoji_start and .emoji_end (code-point offsets), .emoji_rel_position (as emoji_position() defines it, for this occurrence), .emoji_zone (the position of its run, as emoji_runs() reports it), .emoji_prev and .emoji_next ("word", "emoji", "punct" or "none", the nearest thing on that side past any white space), .emoji_prev_space and .emoji_next_space (white space directly on that side), .emoji_intraword, .emoji_run and .emoji_run_length, and .emoji_group, .emoji_subgroup and .emoji_n_codepoints from the catalogue. A grouped input stays grouped.

Details

A "word" is Unicode letters and numbers, and punctuation is anything else that is not white space or an emoji, so nothing here depends on the session's locale. .emoji_intraword marks an emoji with a letter or number directly on both sides, h emoji ow: in ordinary text a strong cue that the emoji stands for a word or a syllable, and the signature of the "Emoji Attack" on language-model judges (Wei et al. 2025), which splits words with emoji to throw off their tokenisation.

References

Na'aman, N., Provenza, H. and Montoya, O. (2017). Varying linguistic purposes of emoji in (Twitter) context. Proceedings of ACL 2017, Student Research Workshop, 136-141. doi:10.18653/v1/P17-3022

Wei, Z. et al. (2025). Emoji attack: enhancing jailbreak attacks against judge LLM detection. Proceedings of ICML 2025, PMLR 267, 66103-66117.

See also

emoji_runs(); emoji_position() for the row-level position.

Examples

d <- data.frame(text = "h\U0001f600ow are you \U0001f600")
emoji_features(d, text)
#> # A tibble: 2 × 18
#>   text             .row_number .emoji_index .emoji .emoji_start .emoji_end
#>   <chr>                  <int>        <int> <chr>         <int>      <int>
#> 1 h😀ow are you 😀           1            1 😀                2          2
#> 2 h😀ow are you 😀           1            2 😀               14         14
#> # ℹ 12 more variables: .emoji_rel_position <dbl>, .emoji_zone <chr>,
#> #   .emoji_prev <chr>, .emoji_next <chr>, .emoji_prev_space <lgl>,
#> #   .emoji_next_space <lgl>, .emoji_intraword <lgl>, .emoji_run <int>,
#> #   .emoji_run_length <int>, .emoji_group <chr>, .emoji_subgroup <chr>,
#> #   .emoji_n_codepoints <int>