最近看代码，发现了一个敏感词检测是用前缀树写的，看起来速度蛮快，毕竟是拿空间换时间，LOG倍速。但是缺点也很明显，待检测文本需要与敏感词词库中的值完全匹配。所以对于简短的词法比较合适。

原理：

1. 每一个节点可以有多个子节点
2. 节点“存储”字符， 节点与节点之间的连线自动形成单词。 如a节点与d节点，之间的连线就是单词 ad
3. 节点可能是叶子节点，此时也是一个单词的“终点”，否则是其他拥有相同前缀的节点的“过客”， wordcount要加一。
4. 删除一个单词，则对应节点上的“过客”都要减一，直至减至叶子节点。

# coding: utf8
MAX_TREE_WIDTH = 26
INIT_CHAR = 'a'
forbiddenwords = """
fuck
fucker
damn
silly
"""
class TrieNode(object):
    def __init__(self):
        self.nodes = [None] * MAX_TREE_WIDTH
        self.wordcount = 0
        self.isend = 0

class TrieTree(object):
    def __init__(self):
        self.root = TrieNode()

    def add(self, word):
        word = word.lower()
        curnode = self.root
        for char in word:
            index = ord(char) - ord(INIT_CHAR)
            if curnode.nodes[index] is None:
                curnode.nodes[index] = TrieNode()
            curnode = curnode.nodes[index]
            curnode.wordcount += 1
        curnode.isend = 1
    
    def search(self, word):
        word = word.lower()
        curnode = self.root
        for char in word:
            index = ord(char) - ord(INIT_CHAR)
            if curnode.nodes[index] is None:
                return -1
            curnode = curnode.nodes[index]
        return curnode.wordcount
    
    def countstartswith(self, prefix):
        curnode = self.root
        for char in prefix:
            index = ord(char) - ord(INIT_CHAR)
            if curnode.nodes[index] is None:
                return -1
            curnode = curnode.nodes[index]
        return curnode.wordcount
    
    def delete(self, word):
        if self.countstartswith(word) > 0:
            curnode = self.root
            for char in word:
                index = ord(char) - ord(INIT_CHAR)
                curnode.nodes[index].wordcount -= 1
                if curnode.nodes[index].wordcount == 0:
                    curnode.nodes[index] = None
                    return
                curnode = curnode.nodes[index]
            curnode.isend = 0

if __name__ == "__main__":
    print("hello trie tree.")
    tree = TrieTree()
    tree.add("h")
    tree.add("He")
    tree.add("hEl")
    tree.add("helL")
    tree.add("hello")
     print(tree.search('he'))
     print(tree.countstartswith("h"))
    print(tree.countstartswith("hel"))
    tree.delete("hel")
    print(tree.countstartswith("hel"))
    print(tree.countstartswith("guo"))
     words = [item for item in forbiddenwords.strip("").split("\n") if item != '']
     for word in words:
         print("adding word: {}".format(word))
         tree.add(word)
     # test sequence
     teststr = "you a silly mother fucker"
     tests = teststr.split(" ")
     for test in tests:
         print("{} --> {}".format(test, tree.search(test)))

运行结果：

hello trie tree.
4
5
3
2
-1
adding word: fuck
adding word: fucker
adding word: damn
adding word: silly
you --> -1
a --> -1
silly --> 1
mother --> -1
fucker --> 1

相较于之前 基于朴素贝叶斯分类算法 的golang 与PHP结合的例子，效率，准确性，适用性都很逊。这也跟场景息息相关，选对了场景，前缀树会给出一个令人满意的答复的。