r - 从 R 中的文本中提取字符级 n-gram

Question

我有一个包含文本的数据框，我想提取字符级双字母组 (n = 2)，例如“st”、“ac”、“ck”，用于 R 中的每个文本。

我还想统计文本中每个字符级二元组的出现频率。

数据:

df$text

[1] "hy my name is"
[2] "stackover flow is great"
[3] "how are you"

Answer 1

我不太确定您在这里的预期输出。我原以为“stack”的二元组将是“st”、“ta”、“ac”和“ck”，因为这会捕获每个连续的对。

例如，如果您想知道单词“brothers”中有多少个二元组“th”，然后将其拆分为二元组“br”、“ot”、“he”和“rs” "，那么你会得到答案 0，这是错误的。

您可以构建一个函数来获取 所有 个二元组，如下所示:

# This function takes a vector of single characters and creates all the bigrams
# within that vector. For example "s", "t", "a", "c", "k" becomes 
# "st", "ta", "ac", and "ck"

pair_chars <- function(char_vec) {
  all_pairs <- paste0(char_vec[-length(char_vec)], char_vec[-1])
  return(as.vector(all_pairs[nchar(all_pairs) == 2]))
}

# This function splits a single word into a character vector and gets its bigrams

word_bigrams <- function(words){
  unlist(lapply(strsplit(words, ""), pair_chars))
}

# This function splits a string or vector of strings into words and gets their bigrams

string_bigrams <- function(strings){
  unlist(lapply(strsplit(strings, " "), word_bigrams))
}

所以现在我们可以在您的示例上进行测试:

df <- data.frame(text = c("hy my name is", "stackover flow is great", 
                          "how are you"), stringsAsFactors = FALSE)

string_bigrams(df$text)
#>  [1] "hy" "my" "na" "am" "me" "is" "st" "ta" "ac" "ck" "ko" "ov" "ve" "er" "fl"
#> [16] "lo" "ow" "is" "gr" "re" "ea" "at" "ho" "ow" "ar" "re" "yo" "ou"

如果你想统计出现次数，你可以使用 table:

table(string_bigrams(df$text))

#> ac am ar at ck ea er fl gr ho hy is ko lo me my na ou ov ow re st ta ve yo 
#>  1  1  1  1  1  1  1  1  1  1  1  2  1  1  1  1  1  1  1  2  2  1  1  1  1 

但是，如果您要进行大量文本挖掘，您应该查看特定的 R 包，例如 stringi、stringr、tm 和 quanteda 帮助完成基本任务

例如，我上面编写的所有基本 R 函数都可以使用 quanteda 包替换，如下所示:

library(quanteda)
char_ngrams(unlist(tokens(df$text, "character")), concatenator = "")
#>  [1] "hy" "ym" "my" "yn" "na" "am" "me" "ei" "is" "ss" "st" "ta" "ac" "ck" 
#> [15] "ko" "ov" "ve" "er" "rf" "fl" "lo" "ow" "wi" "is" "sg" "gr" "re" "ea"
#> [29] "at" "th" "ho" "ow" "wa" "ar" "re" "ey" "yo" "ou"

^{由 reprex package 于 2020-06-13 创建(v0.3.0)}