标注策略：Doccano文本标注

import jsonfrom dataclasses import dataclass, fieldfrom typing import List, Dict, Optional, Tuplefrom collections import defaultdict  @dataclassclass Entity:    start: int    end: int    label: str    text: str = ""  @dataclassclass AnnotationSample:    sample_id: str    text: str    entities: List[Entity] = field(default_factory=list)    annotator: str = ""    approved: bool = False  @dataclassclass DoccanoProject:    """模拟 Doccano 标注项目的核心数据结构与导出功能"""    name: str    project_type: str    labels: List[str] = field(default_factory=list)    relation_types: List[str] = field(default_factory=list)    samples: List[AnnotationSample] = field(default_factory=list)     def add_sample(self, text: str, sample_id: str = "") -> AnnotationSample:        if not sample_id:            sample_id = f"doc-{len(self.samples)+1:04d}"        sample = AnnotationSample(sample_id=sample_id, text=text)        self.samples.append(sample)        return sample     def add_entity(self, sample_id: str, start: int, end: int, label: str):        for s in self.samples:            if s.sample_id == sample_id:                entity_text = s.text[start:end]                s.entities.append(Entity(start=start, end=end, label=label, text=entity_text))                return        raise ValueError(f"Sample {sample_id} not found")     def to_jsonl(self) -> List[dict]:        """导出为 JSONL 格式（Doccano 原生格式）"""        result = []        for s in self.samples:            result.append({                "id": s.sample_id,                "text": s.text,                "entities": [                    {"start": e.start, "end": e.end, "label": e.label, "text": e.text}                    for e in s.entities                ],                "annotator": s.annotator,                "approved": s.approved,            })        return result     def to_spacy(self) -> List[dict]:        """导出为 spaCy 训练格式"""        result = []        for s in self.samples:            entities = [(e.start, e.end, e.label) for e in s.entities]            result.append({                "text": s.text,                "entities": entities,            })        return result     def to_huggingface(self) -> dict:        """导出为 HuggingFace datasets 格式"""        return {            "data": [                {"text": s.text,                 "labels": [e.label for e in s.entities],                 "entities": [                     {"start": e.start, "end": e.end, "label": e.label, "text": e.text}                     for e in s.entities                 ]}                for s in self.samples            ]        }     def statistics(self) -> dict:        """标注进度与标签分布统计（模拟仪表盘）"""        total = len(self.samples)        annotated = sum(1 for s in self.samples if s.entities)        approved = sum(1 for s in self.samples if s.approved)        label_dist = defaultdict(int)        for s in self.samples:            for e in s.entities:                label_dist[e.label] += 1        return {            "total_samples": total,            "annotated": annotated,            "approved": approved,            "progress": round(annotated / max(1, total) * 100, 1),            "label_distribution": dict(label_dist),        }  def calculate_iou(ent1: Entity, ent2: Entity) -> float:    """计算两个实体的交并比(IoU)用于一致性评估"""    intersection_start = max(ent1.start, ent2.start)    intersection_end = min(ent1.end, ent2.end)    intersection = max(0, intersection_end - intersection_start)    union_start = min(ent1.start, ent2.start)    union_end = max(ent1.end, ent2.end)    union = max(1, union_end - union_start)    return round(intersection / union, 4)  def inter_annotator_agreement(samples_a: List[AnnotationSample],                               samples_b: List[AnnotationSample]) -> dict:    """计算两位标注员的标注一致性（标签级+边界级）"""    dict_b = {s.sample_id: s for s in samples_b}    label_match = 0    boundary_iou_sum = 0.0    total_entities = 0     for sa in samples_a:        if sa.sample_id not in dict_b:            continue        sb = dict_b[sa.sample_id]        for ea in sa.entities:            total_entities += 1            best_iou = 0.0            best_label_match = False            for eb in sb.entities:                iou = calculate_iou(ea, eb)                if iou > best_iou:                    best_iou = iou                    best_label_match = (ea.label == eb.label)            if best_label_match and best_iou >= 0.5:                label_match += 1            boundary_iou_sum += best_iou     return {        "total_entities": total_entities,        "matched": label_match,        "label_agreement": round(label_match / max(1, total_entities), 4),        "avg_boundary_iou": round(boundary_iou_sum / max(1, total_entities), 4),    }  # === 演示 ===if __name__ == "__main__":    project = DoccanoProject(        name="合同NER标注项目",        project_type="sequence_labeling",        labels=["当事人", "金额", "日期", "违约条款"],    )     s1 = project.add_sample("甲方张三与乙方公司于2024年1月签订合同，违约金50万元。")    project.add_entity(s1.sample_id, 2, 4, "当事人")    project.add_entity(s1.sample_id, 6, 8, "当事人")    project.add_entity(s1.sample_id, 10, 20, "日期")    project.add_entity(s1.sample_id, 24, 29, "违约条款")    project.add_entity(s1.sample_id, 29, 33, "金额")    s1.annotator = "annotator_A"    s1.approved = True     s2 = project.add_sample("丙方李四应于2024年3月前支付100万元违约金。")    project.add_entity(s2.sample_id, 2, 4, "当事人")    project.add_entity(s2.sample_id, 6, 16, "日期")    project.add_entity(s2.sample_id, 18, 23, "金额")    s2.annotator = "annotator_A"     print("=== 项目统计 ===")    print(json.dumps(project.statistics(), indent=2, ensure_ascii=False))     print("\n=== JSONL导出（前1条） ===")    print(json.dumps(project.to_jsonl()[0], indent=2, ensure_ascii=False))     print("\n=== spaCy格式导出（前1条） ===")    print(json.dumps(project.to_spacy()[0], indent=2, ensure_ascii=False))