从深度嵌套列表提取指定元素并转换为R语言DataFrame

问题场景 #

通过fromJSON()读取JSON文件得到一个深度嵌套列表（部分层级实际为数据框），需要从多个层级提取指定元素，将顶层元素（如PhraseText）与深层元素（如CandidateCUI）关联，生成目标DataFrame。

示例嵌套列表 #

conditions <- list(
  list(
    PMID = 00001,
    Phrases = list(
      list(
        PhraseText = "Hodgkin Lymphoma",
        Mappings = mappings1 <- list(
          list(
            MappingScore = 1000,
            MappingCandidates = mc1 <- list(
              list(CandidateScore = 1000,
                   CandidateCUI = "C075655",
                   CandidateMatched = "Hodgkins Lymphoma",
                   CandidatePreferred = "Hodgkins Lymphoma",
                   MatchedWords = list(c("hodgkin", "lymphoma"))),
              list(CandidateScore = 850,
                   CandidateCUI = "C095659",
                   CandidateMatched = "Lymphoma",
                   CandidatePreferred = "Lymphoma",
                   MatchedWords = list(c("lymphoma"))))
          )
        )
      )
    )
  ),
  list(
    PMID = 00002,
    Phrases = list(
      list(
        PhraseText = "Plaque Psoriasis",
        Mappings = mappings2 <- list(
          list(MappingScore = 1000,
               MappingCandidates = mc2 <- list(
                 list(CandidateScore = 1000,
                      CandidateCUI = "C0125609",
                      CandidateMatched = "Plaque Psoriasis",
                      CandidatePreferred = "Plaque Psoriasis",
                      MatchedWords = list(c("plaque", "psoriasis"))),
                 list(CandidateScore = 750,
                      CandidateCUI = "C0320011",
                      CandidateMatched = "Psoriasis",
                      CandidatePreferred = "Psoriasis",
                      MatchedWords = list(c("psoriasis"))))) 
        )
      )
    )
  )
)

目标DataFrame #

output <- data.frame(
  PhraseText = c("Hodgkin Lymphoma", "Hodgkin Lymphoma", "Plaque Psoriasis", "Plaque Psoriasis"),
  MappingScore = c(1000, 1000, 1000, 1000),
  CandidateScore = c(1000, 850, 1000, 750),
  CandidateCUI = c("C075655", "C095659", "C0125609", "C0320011"),
  CandidatePreferred = c("Hodgkins Lymphoma", "Lymphoma", "Plaque Psoriasis", "Psoriasis")
)

用户尝试代码（存在关联问题） #

x <- lapply(conditions, function(i) {
  lapply(i[["Phrases"]][[1]][["Mappings"]], function(j) {
    lapply(j[["MappingCandidates"]], function(k) {
      k[c("CandidateScore", "CandidateCUI", "CandidatePreferred")]
    })
  })
})

解决方案 #

方法1：使用purrr的map_dfr逐层绑定 #

通过三层嵌套的map_dfr，将上层的PhraseText、MappingScore传递到深层循环中，自动合并所有结果为DataFrame，完美关联各层级元素：

library(tidyverse)

result <- map_dfr(conditions, function(cond) {
  # 提取当前条目下的PhraseText
  phrase_text <- cond$Phrases[[1]]$PhraseText
  # 遍历每个Mapping对象
  map_dfr(cond$Phrases[[1]]$Mappings, function(map) {
    mapping_score <- map$MappingScore
    # 遍历每个候选对象，绑定上层字段
    map_dfr(map$MappingCandidates, function(cand) {
      tibble(
        PhraseText = phrase_text,
        MappingScore = mapping_score,
        CandidateScore = cand$CandidateScore,
        CandidateCUI = cand$CandidateCUI,
        CandidatePreferred = cand$CandidatePreferred
      )
    })
  })
})

# 转为data.frame格式（若需要）
result <- as.data.frame(result)

方法2：利用unnest系列函数扁平化结构（适合含数据框的层级） #

如果嵌套结构中部分层级是数据框，使用tidyverse的unnest_wider和unnest可以更简洁地展开嵌套结构，直接提取所需字段：

library(tidyverse)

result <- tibble(data = conditions) %>%
  # 展开第一层列表为列
  unnest_wider(data) %>%
  # 展开Phrases下的嵌套内容
  unnest_wider(Phrases) %>%
  # 展开Mappings下的嵌套内容
  unnest_wider(Mappings) %>%
  # 展开MappingCandidates列表为多行
  unnest(MappingCandidates) %>%
  # 筛选需要的字段
  select(PhraseText, MappingScore, CandidateScore, CandidateCUI, CandidatePreferred)

# 转为data.frame格式（若需要）
result <- as.data.frame(result)

两种方法都能生成符合要求的目标DataFrame，方法1更通用适配纯列表结构，方法2在存在数据框层级时更高效简洁。

内容的提问来源于stack exchange，提问作者mark_ediger