For a large database, I would like to find a solution where I could predefine the strings to be searched and then get a table that would contain the frequency of these strings and their possible variations per row.
strings <- c("dog", "cat", "mouse")
var1 <- c("black dog", "white dog", "angry dog", "dogs and cats are nice", "dog")
var2 <- c("white cat", "black cat", "tiny cat", NA, "cow")
var3 <- c("little mouse", "big mouse", NA, NA, "mouse")
data <- data.frame(var1, var2, var3)
The result should look like this while I am looking for dog, cat and mouse:
dog&cat 4
mouse 3
We may try
v1 <- do.call(paste, data)
stack(setNames(lapply(c( "\\bdog.*\\bcat|\\bcat.*\\bdog", "mouse"),
\(pat) sum(grepl(pat, v1))), c("dog&cat", "mouse")))[2:1]
ind values
1 dog&cat 4
2 mouse 3
Or if we need all the combinations
lst1 <- lapply(c(strings, combn(strings, 2, FUN = \(x)
sprintf("\\b%1$s.*\\b%2$s|\\b%2$s.*\\b%1$s", x[1], x[2]))),
\(pat) sum(grepl(pat, v1)))
names(lst1) <- c(strings, combn(strings, 2, FUN = paste, collapse = "&"))
stack(lst1)[2:1]
ind values
1 dog 5
2 cat 4
3 mouse 3
4 dog&cat 4
5 dog&mouse 3
6 cat&mouse 2
For more combinations, it may be better to use Reduce with individually applying grepl
lst1 <- lapply(1:3, \(n) {
vals <- colSums(combn(strings, n,
FUN = \(pats) Reduce(`&`, lapply(pats, \(pat) grepl(pat, v1)))))
nms <- combn(strings, n, FUN = paste, collapse = "&")
setNames(vals, nms)
})
stack(unlist(lst1))[2:1]
ind values
1 dog 5
2 cat 4
3 mouse 3
4 dog&cat 4
5 dog&mouse 3
6 cat&mouse 2
7 dog&cat&mouse 2
Or with tidyverse
library(dplyr)
library(stringr)
library(tidyr)
data %>%
unite(var, everything(), na.rm = TRUE, sep = " ") %>%
summarise(`dog&cat` = sum(str_detect(var,
"\\bdog.*\\bcat|\\bcat.*\\bdog")),
mouse = sum(str_detect(var, 'mouse'))) %>%
pivot_longer(everything())
-output
# A tibble: 2 × 2
name value
<chr> <int>
1 dog&cat 4
2 mouse 3