jsonlite::stream_in可以很好地处理这个问题。

library(dplyr)
jsonlite::stream_in(textConnection(df$response), simplifyDataFrame = FALSE) %>%
  lapply(bind_cols) %>%
  bind_rows() %>%
  bind_cols(df[,1,drop=FALSE], .)
#  Imported 2 records. Simplifying...
#            subject aktiv bekümmert interessiert freudig erregt verärgert stark
# 1 dtv85251vucquc45     2         3            4              2         2     0
# 2 mcj8vdqz7sxmjcr0     1         3            1              1         0     0
#   schuldig erschrocken feindselig angeregt stolz gereizt begeistert beschämt
# 1        1           1          1        2     1       1          2        2
# 2        2           0          0        1     2       0          1        0
#   wach nervös entschlossen ängstlich aufmerksam durcheinander
# 1    1      1            1         1          2             2
# 2    2      0            0         0          2             0

或者稍微重构：

library(dplyr)
library(tidyr) # unnest
df %>%
  mutate(
    response = jsonlite::stream_in(textConnection(response), simplifyDataFrame = FALSE),
    response = lapply(response, bind_cols)
  ) %>%
  unnest(response)
#  Imported 2 records. Simplifying...
# # A tibble: 2 × 21
#   subject      aktiv bekümmert interessiert `freudig erregt` verärgert stark schuldig erschrocken feindselig angeregt
#   <chr>        <int>     <int>        <int>            <int>     <int> <int>    <int>       <int>      <int>    <int>
# 1 dtv85251vuc…     2         3            4                2         2     0        1           1          1        2
# 2 mcj8vdqz7sx…     1         3            1                1         0     0        2           0          0        1
# # ℹ 10 more variables: stolz <int>, gereizt <int>, begeistert <int>, beschämt <int>, wach <int>, nervös <int>,
# #   entschlossen <int>, ängstlich <int>, aufmerksam <int>, durcheinander <int>

我不知道“聪明”，但{rjsoncons}实现了“jmespath”查询。我写了一个小助手函数，通过jmespath从json转换到R

from_jmespath = function(json, path) {
    rjsoncons::jmespath(json, path) |>
        jsonlite::fromJSON()
}

然后从数据框的行中提取相关信息

keys = from_jmespath(df$response[1], "*.keys(@)[]")
values = lapply(df$response, from_jmespath, "*.values(@)[]")

然后，我用各种组件构造了一个整洁的tibble

tbl = dplyr::tibble(
    subject = rep(df$subject, each = length(keys)),
    key = rep(keys, nrow(df)),
    value = unlist(values)
)

一个“宽”版本是

tbl |>
    tidyr::pivot_wider(names_from = "key", values_from = "value")