一、引言在互联网内容审核、社交平台监管、评论系统过滤等场景中敏感词检测是一项必不可少的功能。Python凭借其丰富的生态和简洁的语法提供了多种实现敏感词检测的方案。本文将详细介绍几种主流的实现方式并分析各自的优缺点及适用场景。二、基础方案关键词匹配2.1 直接遍历匹配最简单的实现方式是使用列表存储敏感词然后遍历检测pythonclass SimpleSensitiveWordFilter: def __init__(self): self.sensitive_words [] def add_words(self, words): self.sensitive_words.extend(words) def contains_sensitive_word(self, text): for word in self.sensitive_words: if word in text: return True return False def replace_sensitive_words(self, text, replace_char*): result text for word in self.sensitive_words: result result.replace(word, replace_char * len(word)) return result优点实现简单易于理解缺点效率低时间复杂度O(n*m)无法处理重叠匹配2.2 正则表达式匹配使用正则表达式可以更灵活地匹配敏感词pythonimport re class RegexSensitiveWordFilter: def __init__(self): self.pattern None def add_words(self, words): # 使用正则的 | 连接所有敏感词 pattern_str |.join(re.escape(word) for word in words) self.pattern re.compile(pattern_str, re.IGNORECASE) def contains_sensitive_word(self, text): return bool(self.pattern.search(text)) def find_all_sensitive_words(self, text): return self.pattern.findall(text) def replace_sensitive_words(self, text, replace_char*): def replace_func(match): return replace_char * len(match.group()) return self.pattern.sub(replace_func, text)优点支持复杂匹配规则代码简洁缺点敏感词数量大时性能下降正则表达式编译开销较大三、进阶方案前缀树Trie树3.1 原理介绍Trie树前缀树是一种专门用于字符串快速匹配的树形数据结构。它的核心思想是利用字符串的公共前缀来减少不必要的比较实现O(n)的时间复杂度。3.2 实现代码pythonclass TrieNode: __slots__ (children, is_end) def __init__(self): self.children {} self.is_end False class TrieSensitiveWordFilter: def __init__(self): self.root TrieNode() def add_word(self, word): 添加单个敏感词 node self.root for char in word: if char not in node.children: node.children[char] TrieNode() node node.children[char] node.is_end True def add_words(self, words): 批量添加敏感词 for word in words: self.add_word(word) def contains_sensitive_word(self, text): 检测是否包含敏感词 for i in range(len(text)): node self.root j i while j len(text) and text[j] in node.children: node node.children[text[j]] if node.is_end: return True j 1 return False def find_all_sensitive_words(self, text): 找出所有敏感词及位置 results [] for i in range(len(text)): node self.root j i while j len(text) and text[j] in node.children: node node.children[text[j]] if node.is_end: results.append({ word: text[i:j1], start: i, end: j }) j 1 return results def replace_sensitive_words(self, text, replace_char*): 替换敏感词 sensitive_positions self.find_all_sensitive_words(text) if not sensitive_positions: return text result list(text) for pos in sensitive_positions: for k in range(pos[start], pos[end] 1): result[k] replace_char return .join(result)3.3 优化版AC自动机AC自动机Aho-Corasick automaton在Trie树的基础上增加了失败指针实现多模式串的高效匹配pythonfrom collections import deque class AhoCorasickNode: __slots__ (children, fail, output) def __init__(self): self.children {} self.fail None self.output [] # 存储以此节点结尾的敏感词 class AhoCorasickFilter: def __init__(self): self.root AhoCorasickNode() self._built False def add_word(self, word): 添加敏感词 node self.root for char in word: if char not in node.children: node.children[char] AhoCorasickNode() node node.children[char] node.output.append(word) self._built False def add_words(self, words): for word in words: self.add_word(word) def _build_fail_pointers(self): 构建失败指针BFS queue deque() # 初始化第一层节点的失败指针 for char, child in self.root.children.items(): child.fail self.root queue.append(child) while queue: current queue.popleft() for char, child in current.children.items(): queue.append(child) # 寻找失败指针 fail_node current.fail while fail_node is not None and char not in fail_node.children: fail_node fail_node.fail if fail_node is None: child.fail self.root else: child.fail fail_node.children[char] # 合并输出 child.output.extend(child.fail.output) def search(self, text): 搜索文本中的所有敏感词 if not self._built: self._build_fail_pointers() self._built True result [] node self.root for i, char in enumerate(text): # 沿着失败指针查找匹配 while node is not self.root and char not in node.children: node node.fail if char in node.children: node node.children[char] else: node self.root # 收集匹配结果 for word in node.output: result.append({ word: word, position: i - len(word) 1 }) return result def contains_sensitive_word(self, text): return len(self.search(text)) 0 def replace_sensitive_words(self, text, replace_char*): matches self.search(text) if not matches: return text result list(text) for match in matches: start match[position] end start len(match[word]) for i in range(start, end): result[i] replace_char return .join(result)四、第三方库方案4.1 使用better_profanitypythonfrom better_profanity import profanity # 初始化 profanity.load_censor_words() # 检测 text You are a fool if profanity.contains_profanity(text): print(包含敏感词) # 替换 censored_text profanity.censor(text) print(censored_text) # You are a **** # 自定义敏感词库 custom_badwords [badword1, badword2] profanity.load_censor_words(custom_badwords)4.2 使用ahocorasick库pythonimport ahocorasick class PyAhocorasickFilter: def __init__(self): self.automaton ahocorasick.Automaton() self._built False def add_words(self, words): for idx, word in enumerate(words): self.automaton.add_word(word, (idx, word)) self._built False def build(self): self.automaton.make_automaton() self._built True def search(self, text): if not self._built: self.build() result [] for end_index, (idx, word) in self.automaton.iter(text): start_index end_index - len(word) 1 result.append({ word: word, start: start_index, end: end_index }) return result4.3 其他相关库库名特点适用场景profanity-check基于机器学习需要语义理解alt-profanity-check轻量级简单场景ngram-profanity支持模糊匹配变形词检测五、高级特性实现5.1 支持跳过干扰字符pythonclass AdvancedFilter: def __init__(self, skip_charsNone): self.skip_chars skip_chars or { , -, _, .} self.trie_filter TrieSensitiveWordFilter() def _normalize(self, text): 标准化文本去除干扰字符 return .join(c for c in text if c not in self.skip_chars) def contains_sensitive_word(self, text): normalized self._normalize(text) return self.trie_filter.contains_sensitive_word(normalized)5.2 支持拼音/谐音检测python# 使用 pypinyin 库处理拼音 from pypinyin import lazy_pinyin class PinyinSensitiveFilter: def __init__(self): self.pinyin_map {} def add_word(self, word): # 同时添加中文和拼音形式的敏感词 self.trie_filter.add_word(word) pinyin .join(lazy_pinyin(word)) self.trie_filter.add_word(pinyin)5.3 支持英文大小写/变体处理pythonclass CaseInsensitiveFilter: def __init__(self): self.trie_filter TrieSensitiveWordFilter() def add_word(self, word): # 添加小写版本 self.trie_filter.add_word(word.lower()) def contains_sensitive_word(self, text): return self.trie_filter.contains_sensitive_word(text.lower())六、性能对比方案时间复杂度空间复杂度1000词响应时间适用场景直接遍历O(n*m)O(m)~50ms小词库正则表达式O(n*m)O(m)~30ms中等词库Trie树O(n)O(m*k)~5ms大词库AC自动机O(n)O(m*k)~3ms超大词库better_profanityO(n)O(m)~2ms通用场景注n为文本长度m为敏感词数量k为平均词长七、完整实战示例pythonclass SensitiveWordDetectionSystem: 完整的敏感词检测系统 def __init__(self, word_fileNone, skip_charsNone): self.filter AhoCorasickFilter() self.skip_chars skip_chars or { , ., -, _, *, #} if word_file: self.load_words_from_file(word_file) def load_words_from_file(self, file_path): 从文件加载敏感词库 with open(file_path, r, encodingutf-8) as f: words [line.strip() for line in f if line.strip()] self.filter.add_words(words) def preprocess_text(self, text): 预处理文本移除干扰字符 # 可选移除干扰字符 # text .join(c for c in text if c not in self.skip_chars) # 可选转小写 text text.lower() return text def detect(self, text): 检测文本 processed self.preprocess_text(text) return self.filter.search(processed) def audit(self, text, strictTrue): 内容审核 matches self.detect(text) result { is_safe: len(matches) 0, sensitive_words: [m[word] for m in matches], count: len(matches), suggested_action: block if strict and len(matches) 0 else review } return result def censor(self, text, replace_char*): 敏感词脱敏 processed self.preprocess_text(text) matches self.filter.search(processed) if not matches: return text # 需要映射回原始文本的位置 result list(text) # 简化版直接替换实际应用需要处理位置映射 for match in matches: # 这里简化处理 pass return self.filter.replace_sensitive_words(processed, replace_char) # 使用示例 if __name__ __main__: # 初始化系统 system SensitiveWordDetectionSystem() system.filter.add_words([敏感词, 不良信息, 违规内容]) # 测试文本 test_texts [ 这是一条正常的消息, 这里包含敏感词需要处理, 用户发送了不良信息内容 ] for text in test_texts: result system.audit(text) print(f文本: {text}) print(f审核结果: {result}) print(- * 50)八、敏感词库管理建议8.1 词库分类textsensitive_words/ ├── political.txt # 政治敏感 ├── porn.txt # 色情低俗 ├── violence.txt # 暴力恐怖 ├── abuse.txt # 人身攻击 └── spam.txt # 垃圾广告8.2 词库维护策略定期更新建立词库更新机制分级管理按敏感程度分级处理白名单机制支持误杀词的快速恢复版本控制记录词库变更历史九、总结与建议9.1 方案选择指南小型项目100词使用正则或直接遍历即可中型项目100-10000词推荐Trie树方案大型项目10000词必须使用AC自动机快速开发使用better_profanity等成熟库9.2 注意事项性能优化使用缓存、异步处理、批量检测误杀控制建立白名单和人工审核机制编码问题统一使用UTF-8编码变形词需要考虑拼音、谐音、数字替代等多语言中英文混合检测需要特殊处理9.3 扩展方向结合NLP语义理解减少误杀支持图片中的文字检测OCR实时监控和告警系统用户行为分析和分级通过合理选择和优化敏感词检测方案可以有效保障平台内容安全同时保持良好的用户体验。