excel - 一种更准确、更高效的模糊搜索算法

Question

我一直在研究互联网上的模糊匹配/搜索算法。我尝试了几种解决方案。

唯一给出比较准确结果的是 Excel 先生 (http://www.mrexcel.com/pc07.shtml)。这种方法的问题在于单词中字符的顺序或相对位置，而单词本身的顺序对结果没有影响。

我想根据单词的相对位置以及每个单词的字母顺序获得更好的结果。

Function FuzzyMatchByWord(ByVal lsPhrase1 As String, ByVal lsPhrase2 As String, Optional lbStripVowels As Boolean = False, Optional lbDiscardExtra As Boolean = False) As Double

'
' Compare two phrases and return a similarity value (between 0 and 100).
'
' Arguments:
'
' 1. Phrase1        String; any text string
' 2. Phrase2        String; any text string
' 3. StripVowels    Optional to strip all vowels from the phrases
' 4. DiscardExtra   Optional to discard any unmatched words
'


'local variables
Dim lsWord1() As String
Dim lsWord2() As String
Dim ldMatch() As Double
Dim ldCur As Double
Dim ldMax As Double
Dim liCnt1 As Integer
Dim liCnt2 As Integer
Dim liCnt3 As Integer
Dim lbMatched() As Boolean
Dim lsNew As String
Dim lsChr As String
Dim lsKeep As String

'set default value as failure
FuzzyMatchByWord = 0

'create list of characters to keep
lsKeep = "BCDFGHJKLMNPQRSTVWXYZ0123456789 "
If Not lbStripVowels Then
    lsKeep = lsKeep & "AEIOU"
End If

'clean up phrases by stripping undesired characters
'phrase1
lsPhrase1 = Trim$(UCase$(lsPhrase1))
lsNew = ""
For liCnt1 = 1 To Len(lsPhrase1)
    lsChr = Mid$(lsPhrase1, liCnt1, 1)
    If InStr(lsKeep, lsChr) <> 0 Then
        lsNew = lsNew & lsChr
    End If
Next
lsPhrase1 = lsNew
lsPhrase1 = Replace(lsPhrase1, "  ", " ")
lsWord1 = Split(lsPhrase1, " ")
If UBound(lsWord1) = -1 Then
    Exit Function
End If
ReDim ldMatch(UBound(lsWord1))
'phrase2
lsPhrase2 = Trim$(UCase$(lsPhrase2))
lsNew = ""
For liCnt1 = 1 To Len(lsPhrase2)
    lsChr = Mid$(lsPhrase2, liCnt1, 1)
    If InStr(lsKeep, lsChr) <> 0 Then
        lsNew = lsNew & lsChr
    End If
Next
lsPhrase2 = lsNew
lsPhrase2 = Replace(lsPhrase2, "  ", " ")
lsWord2 = Split(lsPhrase2, " ")
If UBound(lsWord2) = -1 Then
    Exit Function
End If
ReDim lbMatched(UBound(lsWord2))

'exit if empty
If Trim$(lsPhrase1) = "" Or Trim$(lsPhrase2) = "" Then
    Exit Function
End If

'compare words in each phrase
For liCnt1 = 0 To UBound(lsWord1)
    ldMax = 0
    For liCnt2 = 0 To UBound(lsWord2)
        If Not lbMatched(liCnt2) Then
            ldCur = FuzzyMatch(lsWord1(liCnt1), lsWord2(liCnt2))
            If ldCur > ldMax Then
                liCnt3 = liCnt2
                ldMax = ldCur
            End If
        End If
    Next
    lbMatched(liCnt3) = True
    ldMatch(liCnt1) = ldMax
Next

'discard extra words
ldMax = 0
For liCnt1 = 0 To UBound(ldMatch)
    ldMax = ldMax + ldMatch(liCnt1)
Next
If lbDiscardExtra Then
    liCnt2 = 0
    For liCnt1 = 0 To UBound(lbMatched)
        If lbMatched(liCnt1) Then
            liCnt2 = liCnt2 + 1
        End If
    Next
Else
    liCnt2 = UBound(lsWord2) + 1
End If

'return overall similarity
FuzzyMatchByWord = 100 * (ldMax / liCnt2)


End Function

Function FuzzyMatch(Fstr As String, Sstr As String) As Double

'
' Code sourced from: http://www.mrexcel.com/pc07.shtml
' Credited to: Ed Acosta
' Modified: Joe Stanton
'

Dim L, L1, L2, M, SC, T, R As Integer

L = 0
M = 0
SC = 1

L1 = Len(Fstr)
L2 = Len(Sstr)

Do While L < L1
    L = L + 1
    For T = SC To L1
        If Mid$(Sstr, L, 1) = Mid$(Fstr, T, 1) Then
            M = M + 1
            SC = T
            T = L1 + 1
        End If
    Next T
Loop

If L1 = 0 Then
    FuzzyMatch = 0
Else
    FuzzyMatch = M / L1
End If

End Function

我正在尝试将试算表中的账户描述与过去 30,000 个账户描述的列表进行比较，我想为每个账户找出前 5 个结果。

举个例子:

Debug.Print FuzzyMatchByWord("Cash and Cash Equivalents", "Bank and Cash")
Debug.Print FuzzyMatchByWord("Cash and Cash Equivalents", "Cash and Bank")
Debug.Print FuzzyMatchByWord("Cash and Cash Equivalents", "Shack sequential")
Debug.Print FuzzyMatchByWord("Cash and Cash Equivalents", "Sequential shack")

返回:

75 
75 
37.5 
37.5

我希望单词在短语中的相对位置对分数的影响更大，我也希望字母的顺序有更大的影响。与现金和现金等价物相比，连续的小屋不应该得分那么高。

Answer 1

比较字符串时，我通常使用 Levenshtein-Distance .您可以找到算法的实现 here .您可以通过比率扩展该函数，这是衡量两个字符串“接近”程度的一个很好的指标。

Function levenshtein(a As String, b As String, Optional ratio As Boolean) As Double

    Dim i As Long, j As Long, cost As Long
    Dim d() As Long
    Dim min1 As Long, min2 As Long, min3 As Long

    If Len(a) = 0 Then
        levenshtein = Len(b)
        Exit Function
    End If

    If Len(b) = 0 Then
        levenshtein = Len(a)
        Exit Function
    End If

    ReDim d(Len(a), Len(b))

    For i = 0 To Len(a)
        d(i, 0) = i
    Next

    For j = 0 To Len(b)
        d(0, j) = j
    Next

    For i = 1 To Len(a)
        For j = 1 To Len(b)
            If Mid(a, i, 1) = Mid(b, j, 1) Then
                cost = 0
            Else
                cost = 1
            End If

            min1 = (d(i - 1, j) + 1)
            min2 = (d(i, j - 1) + 1)
            min3 = (d(i - 1, j - 1) + cost)

            d(i, j) = Application.WorksheetFunction.Min(min1, min2, min3)
        Next
    Next

    If ratio Then
        levenshtein = (Len(a) + Len(b) - d(Len(a), Len(b))) / (Len(a) + Len(b))
    Else
        levenshtein = d(Len(a), Len(b))
    End If

End Function

举个例子:

Debug.Print levenshtein("Cash and Cash Equivalents", "Bank and Cash", True)
Debug.Print levenshtein("Cash and Cash Equivalents", "Cash and Bank", True)
Debug.Print levenshtein("Cash and Cash Equivalents", "Shack sequential", True)
Debug.Print levenshtein("Cash and Cash Equivalents", "Sequential shack", True)

返回:

 0.605263157894737 
 0.631578947368421 
 0.560975609756098 
 0.48780487804878

编辑

我想字符串比较会大大降低速度。加快速度的一种方法是将字符串转换为字节数组并比较数值。这可以这样做:

Function levenshtein(a As String, b As String, Optional ratio As Boolean) As Double

    Dim i As Long, j As Long
    Dim k As Long, l As Long
    Dim cost As Long
    Dim d() As Long
    Dim min1 As Long, min2 As Long, min3 As Long
    Dim aByte1() As Byte, aByte2() As Byte

    If Len(a) = 0 Then
        levenshtein = Len(b)
        Exit Function
    End If

    If Len(b) = 0 Then
        levenshtein = Len(a)
        Exit Function
    End If

    ReDim d(Len(a), Len(b))

    For i = 0 To Len(a)
        d(i, 0) = i
    Next

    For j = 0 To Len(b)
        d(0, j) = j
    Next

    aByte1 = a
    aByte2 = b
    For i = 0 To UBound(aByte1, 1) Step 2
        k = Int(i / 2) + 1
        For j = 0 To UBound(aByte2, 1) Step 2
            If aByte1(i) = aByte2(j) Then
                cost = 0
            Else
                cost = 1
            End If
            l = Int(j / 2) + 1
            min1 = (d(k - 1, l) + 1)
            min2 = (d(k, l - 1) + 1)
            min3 = (d(k - 1, l - 1) + cost)

            d(k, l) = Application.WorksheetFunction.Min(min1, min2, min3)
        Next
    Next

    If ratio Then
        levenshtein = (Len(a) + Len(b) - d(Len(a), Len(b))) / (Len(a) + Len(b))
    Else
        levenshtein = d(Len(a), Len(b))
    End If

End Function