基于DFA的敏感词过滤

在计算理论中,确定有限状态自动机或确定有限自动机(英语:deterministic finite automaton, DFA)是一个能实现状态转移的自动机。对于一个给定的属于该自动机的状态和一个属于该自动机字母表{\displaystyle \Sigma }Σ的字符,它都能根据事先给定的转移函数转移到下一个状态

DFA算法

DFA((Deterministic Finite automation))确定性的有穷状态自动机: 从一个状态输入一个字符集合能到达下一个确定的状态。如图:

 
dfa_1.png

如上图当AB状态输入a得到状态aB,状态aB输入b得到状态ab; 状态AB输入b得到状态Ab,状态Ab输入a得到状态ab。

利用DFA匹配关键词

上面开始的几个关键词匹配可以用下图来表示:

 
dfa_2.png

0是开始状态,输入日、本、人会最终到达结束状态5,输入日、本、鬼、子最终到达结束状态8,输入中、国、人到达结束状态7。

以上的状态图输入字符类似树形结构,空心状态表示未结束状态(isEnd=false), 蓝色环形状态表示结束状态(isEnd=true)。用HashMap维护这个字典关系.

 

代码(Python3 非原创代码):

from collections import defaultdict
import re

__all__ = ['NaiveFilter', 'BSFilter', 'DFAFilter']
__author__ = 'observer'
__update__ = 'obaby@mars https://www.h4ck.org.cn'
__update_date__ = '2019.11.27'


class NaiveFilter():
    '''Filter Messages from keywords

    very simple filter implementation

    >>> f = NaiveFilter()
    >>> f.add("sexy")
    >>> f.filter("hello sexy baby")
    hello **** baby
    '''

    def __init__(self):
        self.keywords = set([])

    def parse(self, path):
        for keyword in open(path):
            self.keywords.add(keyword.strip().decode('utf-8').lower())

    def filter(self, message, repl="*"):
        message = message.lower()
        for kw in self.keywords:
            message = message.replace(kw, repl)
        return message


class BSFilter:
    '''Filter Messages from keywords

    Use Back Sorted Mapping to reduce replacement times

    >>> f = BSFilter()
    >>> f.add("sexy")
    >>> f.filter("hello sexy baby")
    hello **** baby
    '''

    def __init__(self):
        self.keywords = []
        self.kwsets = set([])
        self.bsdict = defaultdict(set)
        self.pat_en = re.compile(r'^[0-9a-zA-Z]+$')  # english phrase or not

    def add(self, keyword):
        if not isinstance(keyword, str):
            keyword = keyword.decode('utf-8')
        keyword = keyword.lower()
        if keyword not in self.kwsets:
            self.keywords.append(keyword)
            self.kwsets.add(keyword)
            index = len(self.keywords) - 1
            for word in keyword.split():
                if self.pat_en.search(word):
                    self.bsdict[word].add(index)
                else:
                    for char in word:
                        self.bsdict[char].add(index)

    def parse(self, path):
        with open(path, "r") as f:
            for keyword in f:
                self.add(keyword.strip())

    def filter(self, message, repl="*"):
        if not isinstance(message, str):
            message = message.decode('utf-8')
        message = message.lower()
        for word in message.split():
            if self.pat_en.search(word):
                for index in self.bsdict[word]:
                    message = message.replace(self.keywords[index], repl)
            else:
                for char in word:
                    for index in self.bsdict[char]:
                        message = message.replace(self.keywords[index], repl)
        return message


class DFAFilter():
    '''Filter Messages from keywords

    Use DFA to keep algorithm perform constantly

    >>> f = DFAFilter()
    >>> f.add("sexy")
    >>> f.filter("hello sexy baby")
    hello **** baby
    '''

    def __init__(self):
        self.keyword_chains = {}
        self.delimit = '\x00'

    def add(self, keyword):
        if not isinstance(keyword, str):
            keyword = keyword.decode('utf-8')
        keyword = keyword.lower()
        chars = keyword.strip()
        if not chars:
            return
        level = self.keyword_chains
        for i in range(len(chars)):
            if chars[i] in level:
                level = level[chars[i]]
            else:
                if not isinstance(level, dict):
                    break
                for j in range(i, len(chars)):
                    level[chars[j]] = {}
                    last_level, last_char = level, chars[j]
                    level = level[chars[j]]
                last_level[last_char] = {self.delimit: 0}
                break
        if i == len(chars) - 1:
            level[self.delimit] = 0

    def parse(self, path):
        with open(path, encoding='UTF-8') as f:
            for keyword in f:
                self.add(keyword.strip())

    def filter(self, message, repl="*"):
        if not isinstance(message, str):
            message = message.decode('utf-8')
        message = message.lower()
        ret = []
        start = 0
        while start < len(message):
            level = self.keyword_chains
            step_ins = 0
            for char in message[start:]:
                if char in level:
                    step_ins += 1
                    if self.delimit not in level[char]:
                        level = level[char]
                    else:
                        ret.append(repl * step_ins)
                        start += step_ins - 1
                        break
                else:
                    ret.append(message[start])
                    break
            else:
                ret.append(message[start])
            start += 1

        return ''.join(ret)

    def is_contain_sensi_key_word(self, message):
        repl = '_-__-'
        dest_string = self.filter(message=message, repl=repl)
        if repl in dest_string:
            return True
        return False


def test_first_character():
    gfw = DFAFilter()
    gfw.add("1989年")
    assert gfw.filter("1989", "*") == "1989"


if __name__ == "__main__":
    # gfw = NaiveFilter()
    # gfw = BSFilter()
    gfw = DFAFilter()
    gfw.parse("keywords")
    import time

    t = time.process_time()
    print(gfw.filter("法 我操操操", "*"))
    print(gfw.filter("针孔摄像机 我操操操", "*"))
    print(gfw.filter("传世私服 我操操操", "*"))
    print('Cost is %6.6f' % (time.process_time() - t))
    print(gfw.is_contain_sensi_key_word('习大大'))
    test_first_character()

Github: https://github.com/obaby/dfa-python-filter

参考链接:

https://www.jianshu.com/p/e58a148eecc5

https://blog.csdn.net/u013421629/article/details/83178970

https://www.jianshu.com/p/c0c7667dc6c5

https://blog.csdn.net/xiqi4145/article/details/84313809


分享文章:

猜你喜欢:

发表评论

您的电子邮箱地址不会被公开。 必填项已用*标注