Finding all terms near a given phrase - r

Question

Is there a way to find all words that are associated with a given phrase?
For example, say I want to find all of the words next to the word "illness" in a string. The string has the word "illness" quite a bit, and I want to find all of the terms surrounding it, such as "has illness," "does not have illness," "might have illness," etc...

Answer 1

If you just have a vector of strings (isolated or as a column in a data.frame), then perhaps:
s <- c("hello world illness quux bar not action illness",
"not an illness", "no positive", "illness", "illness goodbye")
ret <- lapply(list(gregexpr("(\\S+)(?= illness)", s, perl = TRUE),
gregexpr("(?<=illness )(\\S+)", s, perl = TRUE)),
regmatches, x = s)
Map(c, ret[[1]], ret[[2]])
# [[1]]
# [1] "world" "action" "quux"
# [[2]]
# [1] "an"
# [[3]]
# character(0)
# [[4]]
# character(0)
# [[5]]
# [1] "goodbye"
Each gregexpr finds a word (well, contiguous non-whitespace) followed by the literal " illness" or preceded by the literal "illness ". Because it returns a list with enough information to extract the substrings from the original, we use regmatches(x=s, ...) to extract the components.
(The Map command deals with the fact that the lapply result is a list, length 2, one for each regex. It merely concatenates the substrings from the first regex with the substrings from the second regex, "zipping" the matches for each string within the vector. If you look at ret[[1]] and/or ret as a whole, the benefit of this might make more sense.)
If you don't care in which string in the vector/column the surrounding words are found, then you can simply unlist this:
unlist(ret)
# [1] "world" "action" "an" "quux" "goodbye"

Answer 2

Quite unclear what pattern exactly you want to match. So here's some options, using str_extractfrom the stringrpackage and positive lookahead:
dt <- c("I mean he's got to have some illness I suppose that",
"whether you've had a serious illness or [unclear]",
"No not what illnesses, terminal illness.",
"Oh, Terminal illness, oh, sorry.",
"Illness benefit on a joint life, last survivor plan?")
str_extract(dt, ".*(?=.(i|I)llness)") # any string prior to "illness"
[1] "I mean it's got to be some some" "whether you've had a serious" "No not what illnesses, terminal"
[4] "Oh, Terminal" NA
str_extract(dt, "(some|serious|terminal)(?=.(i|I)llness)") # specific words prior to "illness"
[1] "some" "serious" "terminal" NA NA
str_extract(dt, "\\w+\\b(?=.(i|I)llness)") # last word prior to "illness"
[1] "some" "serious" "what" "Terminal" NA

Answer 3

Using the kwic function from quanteda. You can specify the number of words surrounding the term you are looking for. In the example I'm using 3, but that can be anything. The advantage is that you can store the outcome in a data.frame and then do some more investigations.
Using the example from #Chris Ruehlemann:
library(quanteda)
dt <- c("I mean he's got to have some illness I suppose that",
"whether you've had a serious illness or [unclear]",
"No not what illnesses, terminal illness.",
"Oh, Terminal illness, oh, sorry.",
"Illness benefit on a joint life, last survivor plan?")
out <- kwic(dt, pattern = "illness", window = 3, valuetype = "regex")
out
[text1, 8] to have some | illness | I suppose that
[text2, 6] had a serious | illness | or[ unclear
[text3, 4] No not what | illnesses | , terminal illness
[text3, 7] illnesses, terminal | illness | .
[text4, 4] Oh, Terminal | illness | , oh,
[text5, 1] | Illness | benefit on a
data.frame(out)
docname from to pre keyword post pattern
1 text1 8 8 to have some illness I suppose that illness
2 text2 6 6 had a serious illness or [ unclear illness
3 text3 4 4 No not what illnesses , terminal illness illness
4 text3 7 7 illnesses , terminal illness . illness
5 text4 4 4 Oh , Terminal illness , oh , illness
6 text5 1 1 Illness benefit on a illness