How to extract only person A's statements in a conversation between two persons A and B

Question

I have a record of conversations between two arbitrary persons A and B.

c1 <- "Person A: blabla...something Person B: blabla something else Person A: OK blabla"
c2 <- "Person A: again blabla Person B: blabla something else Person A: thanks blabla"

The data frame looks like this:

df <- data.frame(id = rbind(123, 345), conversation = rbind(c1, c2))

df

    id                                                                     conversation
c1 123 Person A: blabla...something Person B: blabla something else Person A: OK blabla
c2 345   Person A: again blabla Person B: blabla something else Person A: thanks blabla

Now I would like to extract only the part of person A and put it in a data frame. The result should be:

   id                     person_A
1 123 blabla...something OK blabla
2 345   again blabla thanks blabla

Answer 1

I'm a big fan of solving this sort of problem in a way that gives you access to all the data (that includes Person B's discourse as well). I love tidyr's extract for this sort of column splitting. I used to use a do.call(rbind, strsplit())) approach but love how clean the extract approach is.

c1 <- "Person A: blabla...something Person B: blabla something else Person A: OK blabla"
c2 <- "Person A: again blabla Person B: blabla something else Person A: thanks blabla"
c3 <- "Person A: again blabla Person B: blabla something else"
df <- data.frame(id = rbind(123, 345, 567), conversation = rbind(c1, c2, c3))


if (!require("pacman")) install.packages("pacman")
pacman::p_load(dplyr, tidyr)

conv <- strsplit(as.character(df[["conversation"]]), "\\s+(?=Person\\s)", perl=TRUE)

df2 <- df[rep(1:nrow(df), sapply(conv, length)), ,drop=FALSE]
rownames(df2) <- NULL
df2[["conversation"]] <- unlist(conv)

df2 %>%
    extract(conversation, c("Person", "Conversation"), "([^:]+):\\s+(.+)")

##    id   Person          Conversation
## 1 123 Person A    blabla...something
## 2 123 Person B blabla something else
## 3 123 Person A             OK blabla
## 4 345 Person A          again blabla
## 5 345 Person B blabla something else
## 6 345 Person A         thanks blabla
## 7 567 Person A          again blabla
## 8 567 Person B blabla something else


df2 %>%
    extract(conversation, c("Person", "Conversation"), "([^:]+):\\s+(.+)") %>%
    filter(Person == "Person A")    

##    id   Person       Conversation
## 1 123 Person A blabla...something
## 2 123 Person A          OK blabla
## 3 345 Person A       again blabla
## 4 345 Person A      thanks blabla
## 5 567 Person A       again blabla

Or collapse them as you show in the desired output:

df2 %>%
    extract(conversation, c("Person", "Conversation"), "([^:]+):\\s+(.+)") %>%
    filter(Person == "Person A") %>%
    group_by(id) %>%
    select(-Person) %>%
    summarise(Person_A =paste(Conversation, collapse=" "))

##    id                     Person_A
## 1 123 blabla...something OK blabla
## 2 345   again blabla thanks blabla
## 3 567                 again blabla

Edit: In reality I suspect your data has real names like "john Smith" vs. "Person A". If this is the case this initial regex split will capture a first and last name that uses caps followed by a colon:

c1 <- "Greg Smith: blabla...something Sue Williams: blabla something else Greg Smith: OK blabla"
c2 <- "Greg Smith: again blabla Sue Williams: blabla something else Greg Smith: thanks blabla"
c3 <- "Greg Smith: again blabla Sue Williams: blabla something else"
df <- data.frame(id = rbind(123, 345, 567), conversation = rbind(c1, c2, c3))r


conv <- strsplit(as.character(df[["conversation"]]), "\\s+(?=([A-Z][a-z]+\\s+[A-Z][a-z]+:))", perl=TRUE)

df2 <- df[rep(1:nrow(df), sapply(conv, length)), ,drop=FALSE]
rownames(df2) <- NULL
df2[["conversation"]] <- unlist(conv)

df2 %>%
    extract(conversation, c("Person", "Conversation"), "([^:]+):\\s+(.+)")

##    id       Person          Conversation
## 1 123   Greg Smith    blabla...something
## 2 123 Sue Williams blabla something else
## 3 123   Greg Smith             OK blabla
## 4 345   Greg Smith          again blabla
## 5 345 Sue Williams blabla something else
## 6 345   Greg Smith         thanks blabla
## 7 567   Greg Smith          again blabla
## 8 567 Sue Williams blabla something else

Answer 2

Using data.table andgsub` from base R:

require(data.table)
setDT(df)[, Person_A := gsub(".*Person A:[ ]*(.*)[ ]*Person B.*:[ ]*(.*)$", 
                         "\\1\\2", conversation)][, conversation := NULL]
df
#     id                       Person_A
# 1: 123 blabla...something OK blabla
# 2: 345   again blabla thanks blabla

Answer 3

This is my try, I have also added a second conversation started by Person B and a conversation also ended by Person B, just to cover also these cases:

c1 <- "Person A: blabla...something Person B: blabla something else Person A: OK blabla"
c2 <- "Person A: again blabla Person B: blabla something else Person A: thanks blabla"
c3 <- "Person A: again blabla Person B: blabla something else"
df <- data.frame(id = rbind(123, 345, 567), conversation = rbind(c1, c2, c3))


df$PersonA <- gsub("(Person A: |Person B: .+? (?<= Person A: )|Person B: .+?\\Z)", "", df$conversation, perl = TRUE)
df$PersonA

What I'm doing with gsub is removing:

1. Person A:
2. Person B's sentences followed by A's sentences
3. B's sentences at the end of the conversaiont \Z

I used the perl = TRUE because life is too short to not to use the rearview mirror... ehm... the lookbehind operator.

Answer 4

Using the stringr package

First we split the string using "Person A: " as a delimiter

library(stringr)
conv.split <- str_split(df$conversation, "Person A: ")

This will give us all pieces of conversation started by A with attached the (optional) answer by B

We now remove B's answers

conv.split <- lapply(conv.split, function(x){str_split(x, "Person B:.*")})

And finally we unlist each element and collapse it together into a string

sapply(conv.split, function(x){x <- unlist(x); paste(x, collapse = "")})

Result:

[1] "blabla...something OK blabla" "again blabla thanks blabla" 

Works also in the case where B starts the conversation, if only one of the two is speaking and also for long conversations.

Answer 5

It might not work for all your cases. Especially ones that the conversation is started from Person B. Let me know if it is the case. Else try

df$person_A <- gsub("Person B.*:|Person A:", "", df$conversation)
df <- data.frame(df$id, df$person_A)