使用 ArangoDB AQL 计算字符串出现次数

Question

要计算包含特定属性值的对象的数量，我可以执行以下操作：

FOR t IN thing
  COLLECT other = t.name = "Other" WITH COUNT INTO otherCount
  FILTER other != false
  RETURN otherCount

但是如何计算同一查询中的其他三个事件，而不会导致子查询多次通过同一数据集运行？

我尝试过类似的方法：

FOR t IN thing
  COLLECT 
    other = t.name = "Other",
    some = t.name = "Some",
    thing = t.name = "Thing"
  WITH COUNT INTO count
  RETURN {
   other, some, thing,
   count
  }

但我无法理解结果：我一定是以错误的方式处理这个问题？

Answer 1

拆分计数

您可以按短语拆分字符串并从计数中减去 1。这适用于任何子字符串，另一方面意味着它不尊重单词边界。

LET things = [
    {name: "Here are SomeSome and Some Other Things, brOther!"},
    {name: "There are no such substrings in here."},
    {name: "some-Other-here-though!"}
]

FOR t IN things
  LET Some = LENGTH(SPLIT(t.name, "Some"))-1
  LET Other = LENGTH(SPLIT(t.name, "Other"))-1
  LET Thing = LENGTH(SPLIT(t.name, "Thing"))-1
  RETURN {
   Some, Other, Thing
}

结果：

[
  {
    "Some": 3,
    "Other": 2,
    "Thing": 1
  },
  {
    "Some": 0,
    "Other": 0,
    "Thing": 0
  },
  {
    "Some": 0,
    "Other": 1,
    "Thing": 0
  }
]

您可以使用SPLIT(LOWER(t.name), LOWER("...")) 使其不区分大小写。

收集单词

TOKENS() 函数可用于将输入拆分为单词数组，然后可以对其进行分组和计数。请注意，我稍微更改了输入。输入 "SomeSome" 将不会被计算在内，因为 "somesome" != "some"（此变体是单词而不是基于子字符串）。

LET things = [
    {name: "Here are SOME some and Some Other Things. More Other!"},
    {name: "There are no such substrings in here."},
    {name: "some-Other-here-though!"}
]
LET whitelist = TOKENS("Some Other Things", "text_en")

FOR t IN things
  LET whitelisted = (FOR w IN TOKENS(t.name, "text_en") FILTER w IN whitelist RETURN w)
  LET counts = MERGE(FOR w IN whitelisted
    COLLECT word = w WITH COUNT INTO count
    RETURN { [word]: count }
  )
  RETURN {
    name: t.name,
    some: counts.some || 0,
    other: counts.other || 0,
    things: counts.things ||0
  }

结果：

[
  {
    "name": "Here are SOME some and Some Other Things. More Other!",
    "some": 3,
    "other": 2,
    "things": 0
  },
  {
    "name": "There are no such substrings in here.",
    "some": 0,
    "other": 0,
    "things": 0
  },
  {
    "name": "some-Other-here-though!",
    "some": 1,
    "other": 1,
    "things": 0
  }
]

这确实使用了 COLLECT 的子查询，否则它将计算整个输入的出现总数。

白名单步骤不是绝对必要的，您也可以让它计算所有单词。对于较大的输入字符串，它可能会节省一些内存，而不是为您不感兴趣的单词执行此操作。

如果您想精确匹配单词，您可能需要为该语言创建一个单独的Analyzer with stemming disabled。您也可以关闭normalization ("accent": true, "case": "none")。另一种方法是将REGEX_SPLIT() 用于典型的空白和标点符号以进行更简单的标记，但这取决于您的用例。

其他解决方案

我认为不可能在没有子查询的情况下使用 COLLECT 独立计算每个输入对象，除非您想要总计数。

拆分有点小技巧，但是您可以将 SPLIT() 替换为 REGEX_SPLIT() 并将搜索短语包装在 \b 中以仅在单词边界位于两侧时才匹配。那么它应该只匹配单词（或多或少）：

LET things = [
    {name: "Here are SomeSome and Some Other Things, brOther!"},
    {name: "There are no such substrings in here."},
    {name: "some-Other-here-though!"}
]

FOR t IN things
  LET Some = LENGTH(REGEX_SPLIT(t.name, "\\bSome\\b"))-1
  LET Other = LENGTH(REGEX_SPLIT(t.name, "\\bOther\\b"))-1
  LET Thing = LENGTH(REGEX_SPLIT(t.name, "\\bThings\\b"))-1
  RETURN {
   Some, Other, Thing
}

结果：

[
  {
    "Some": 1,
    "Other": 1,
    "Thing": 1
  },
  {
    "Some": 0,
    "Other": 0,
    "Thing": 0
  },
  {
    "Some": 0,
    "Other": 1,
    "Thing": 0
  }
]

更优雅的解决方案是使用 ArangoSearch 进行字数统计，但它没有让您检索单词出现频率的功能。它可能已经在内部跟踪了这一点（分析器功能“频率”），但它绝对不会在此时公开。