传统人工核查模拟（伪代码）
def traditional_fact_check(claim):
     步骤1：人工收集相关线索
    clues = manually_search_keywords(claim)
     步骤2：查阅权威文献/数据库
    references = manually_check_references(clues)
     步骤3：咨询领域专家
    expert_opinion = consult_expert(claim, references)
     步骤4：人工综合判断
    verdict = human_judgment(expert_opinion)
    return verdict   耗时：6-8小时

 AI辟谣助手核心工作流示例
 基于RAG + 简易知识图谱模拟

import requests
from typing import Dict, List, Tuple

class SimpleRumorChecker:
    """
    极简AI辟谣助手核心类
    演示 RAG + 知识图谱 的基础工作流程
    """
    
    def __init__(self):
         模拟本地知识图谱（实体-关系-实体）
        self.knowledge_graph = {
            ("中国载人航天", "首飞年份", "2003"),
            ("神舟五号", "航天员", "杨利伟"),
            ("袁隆平", "研究领域", "杂交水稻"),
            ("杂交水稻", "亩产记录", "1530公斤"),
        }
         模拟权威信源库（用于RAG检索）
        self.authority_sources = [
            {"source": "中国政府网", "content": "2025年，中国GDP增长率为5.0%"},
            {"source": "国家统计局", "content": "2025年全国居民人均可支配收入41314元"},
            {"source": "人民日报", "content": "神舟十九号于2025年10月成功发射"},
        ]
    
     --- 步骤1：断言提取 ---
    def extract_claims(self, text: str) -> List[str]:
        """从原始文本中提取待验证的核心断言"""
         简化版：按标点分割，过滤掉提问句和感叹句
        sentences = text.replace("？", "。").replace("！", "。").split("。")
        claims = [s.strip() for s in sentences if len(s.strip()) > 5]
        return claims
    
     --- 步骤2：RAG检索 ---
    def retrieve_evidence(self, claim: str) -> List[Tuple[str, str, float]]:
        """
        检索与声明相关的证据（模拟RAG的检索阶段）
        返回：(证据来源, 证据内容, 相似度得分)
        """
        evidence_list = []
        for source in self.authority_sources:
             模拟关键词匹配检索
            relevance = self._calc_relevance(claim, source["content"])
            if relevance > 0.3:
                evidence_list.append((source["source"], source["content"], relevance))
        return sorted(evidence_list, key=lambda x: x[2], reverse=True)
    
    def _calc_relevance(self, claim: str, content: str) -> float:
        """计算声明与证据的相关性（简化版余弦相似度模拟）"""
        claim_words = set(claim)
        content_words = set(content)
        if not claim_words or not content_words:
            return 0.0
        intersection = claim_words & content_words
        return len(intersection) / (len(claim_words) + len(content_words) - len(intersection))
    
     --- 步骤3：知识图谱验证 ---
    def verify_with_kg(self, claim: str) -> Tuple[bool, str]:
        """利用知识图谱验证声明的真伪"""
        for (entity, relation, value) in self.knowledge_graph:
            if entity in claim and relation in claim:
                 检查声明中提到的值与知识图谱中的值是否一致
                if value.lower() in claim.lower():
                    return True, f"知识图谱验证通过：{entity} {relation} {value}"
                else:
                    return False, f"知识图谱验证冲突：知识图谱记录{entity} {relation} = {value}"
        return None, "知识图谱中未找到相关事实"
    
     --- 步骤4：综合判定与置信度评分 ---
    def check(self, user_input: str) -> Dict:
        """
        主入口：对用户输入的文本进行辟谣检测
        返回：判定结果 + 置信度 + 证据链
        """
         步骤1：提取断言
        claims = self.extract_claims(user_input)
        if not claims:
            return {"verdict": "无法解析", "confidence": 0.0, "evidence": []}
        
        all_evidence = []
        kg_results = []
        
        for claim in claims:
             步骤2：RAG检索证据
            rag_evidence = self.retrieve_evidence(claim)
            all_evidence.extend(rag_evidence)
            
             步骤3：知识图谱验证
            kg_result = self.verify_with_kg(claim)
            if kg_result[0] is not None:
                kg_results.append(kg_result)
        
         步骤4：综合判定
        if kg_results:
             知识图谱优先
            conflict_count = sum(1 for r in kg_results if not r[0])
            if conflict_count > 0:
                verdict = "存在疑点/可能为谣言"
                confidence = max(0.5, 1.0 - conflict_count / len(kg_results))
            else:
                verdict = "信息可信"
                confidence = 0.8
        elif all_evidence:
            verdict = "证据不足，建议进一步核实"
            confidence = 0.4
        else:
            verdict = "无法验证"
            confidence = 0.1
        
        return {
            "verdict": verdict,
            "confidence": round(confidence, 2),
            "evidence": all_evidence[:3],   返回前3条证据
            "kg_check": kg_results
        }


 运行示例
if __name__ == "__main__":
    checker = SimpleRumorChecker()
    
     测试案例1：真消息
    test_true = "中国载人航天首飞于2003年完成。"
    result1 = checker.check(test_true)
    print(f"输入: {test_true}")
    print(f"判定: {result1['verdict']} (置信度: {result1['confidence']})")
    print(f"知识图谱验证: {result1['kg_check']}")
    
    print("\n" + "="50 + "\n")
    
     测试案例2：假消息
    test_false = "神舟五号的航天员是翟志刚。袁隆平研究的是转基因玉米。"
    result2 = checker.check(test_false)
    print(f"输入: {test_false}")
    print(f"判定: {result2['verdict']} (置信度: {result2['confidence']})")
    for kg in result2['kg_check']:
        print(f"   {kg[1]}")
    
    print("\n" + "="50 + "\n")
    
     测试案例3：需要RAG检索的消息
    test_rag = "2025年中国GDP增长率为5.0%。"
    result3 = checker.check(test_rag)
    print(f"输入: {test_rag}")
    print(f"判定: {result3['verdict']} (置信度: {result3['confidence']})")
    for evidence in result3['evidence']:
        print(f"   来源: {evidence[0]}, 内容: {evidence[1]}")