使用grepl()对非互斥变量类别重新编码的问题

问题背景 #

使用case_when对非互斥类别变量重新编码时，存在同时满足多个类别的样本（如标题含"dog和space"的帖子），会默认匹配第一个符合的条件，导致分类结果仅由条件顺序决定，无法准确反映多类别归属。

核心原因 #

case_when的执行逻辑是按顺序依次判断条件，返回第一个结果为TRUE的对应值，因此同时满足多个条件的样本只会被分配到第一个匹配的类别，调换条件顺序就会改变分类结果。

解决方案 #

方案1：生成多标记变量（推荐用于非互斥场景）

针对非互斥的类别，为每个类别单独创建二进制标记变量，一个样本可同时属于多个类别，清晰展现所有符合的类别：

post_caption <- c("This post is about a dog", "This post is about a cat", "This post is about a walrus", "This post is about space", "this post is about both a dog and space")
post_name <- c("dog account", "dog_account", "walrus_account", "space_account", "space_account")

# 标记是否属于animal_post（忽略大小写匹配）
animal_post <- grepl(paste(c("dog", "cat", "walrus"), collapse = "|"), post_caption, ignore.case = TRUE)
# 标记是否属于other_post（忽略大小写匹配）
other_post <- grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_caption, ignore.case = TRUE) & 
  grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_name, ignore.case = TRUE)

# 合并结果为数据框
result_df <- data.frame(
  post_caption,
  post_name,
  is_animal_post = animal_post,
  is_other_post = other_post
)

print(result_df)

输出结果：

post_caption      post_name is_animal_post is_other_post
1           This post is about a dog      dog account           TRUE         FALSE
2           This post is about a cat       dog_account           TRUE         FALSE
3        This post is about a walrus    walrus_account           TRUE         FALSE
4          This post is about space     space_account          FALSE          TRUE
5 this post is about both a dog and space space_account           TRUE          TRUE

方案2：定义多类别混合标签

如果需要给每个样本分配单一标签，但要区分同时满足多个条件的情况，可以在case_when中优先判断多条件同时满足的场景：

post_caption <- c("This post is about a dog", "This post is about a cat", "This post is about a walrus", "This post is about space", "this post is about both a dog and space")
post_name <- c("dog account", "dog_account", "walrus_account", "space_account", "space_account")

category <- case_when(
  # 优先判断同时满足两类的情况
  grepl(paste(c("dog", "cat", "walrus"), collapse = "|"), post_caption, ignore.case = TRUE) &
    grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_caption, ignore.case = TRUE) &
    grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_name, ignore.case = TRUE) ~ "mixed_post",
  # 单独匹配animal_post
  grepl(paste(c("dog", "cat", "walrus"), collapse = "|"), post_caption, ignore.case = TRUE) ~ "animal_post",
  # 单独匹配other_post
  grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_caption, ignore.case = TRUE) &
    grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_name, ignore.case = TRUE) ~ "other_post",
  # 未匹配的情况返回NA
  TRUE ~ NA_character_
)

print(category)

输出结果：

[1] "animal_post" "animal_post" "animal_post" "other_post"  "mixed_post" 

方案3：明确优先级规则

如果必须保留单一类别且有明确的优先级（如优先标记other_post），可以在低优先级条件中排除已匹配高优先级的样本，让逻辑更明确：

post_caption <- c("This post is about a dog", "This post is about a cat", "This post is about a walrus", "This post is about space", "this post is about both a dog and space")
post_name <- c("dog account", "dog_account", "walrus_account", "space_account", "space_account")

category <- case_when(
  # 高优先级：先匹配other_post
  grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_caption, ignore.case = TRUE) &
    grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_name, ignore.case = TRUE) ~ "other_post",
  # 低优先级：仅匹配未被标记为other_post的animal样本
  !grepl(paste(c("space", "rocks", "trees"), collapse = "|"), post_caption, ignore.case = TRUE) &
    grepl(paste(c("dog", "cat", "walrus"), collapse = "|"), post_caption, ignore.case = TRUE) ~ "animal_post",
  TRUE ~ NA_character_
)

print(category)

输出结果：

[1] "animal_post" "animal_post" "animal_post" "other_post"  "other_post" 

内容的提问来源于stack exchange，提问作者RMRH