video-compliance-ai/backend/app/schemas/review.py

"""
审核相关的 Pydantic 模型（API 契约定义）
所有测试和实现必须遵循此契约
"""
from typing import Optional
from datetime import datetime
from pydantic import BaseModel, Field, HttpUrl
from enum import Enum


# ==================== 枚举定义 ====================

class Platform(str, Enum):
    """支持的投放平台"""
    DOUYIN = "douyin"
    XIAOHONGSHU = "xiaohongshu"
    BILIBILI = "bilibili"
    KUAISHOU = "kuaishou"


class TaskStatus(str, Enum):
    """任务状态"""
    PENDING = "pending"
    PROCESSING = "processing"
    COMPLETED = "completed"
    FAILED = "failed"
    APPROVED = "approved"
    REJECTED = "rejected"


class RiskLevel(str, Enum):
    """风险等级"""
    HIGH = "high"      # 法律违规（广告法极限词）
    MEDIUM = "medium"  # 平台规则违规
    LOW = "low"        # 品牌规范违规


class ViolationType(str, Enum):
    """违规类型"""
    FORBIDDEN_WORD = "forbidden_word"    # 违禁词
    EFFICACY_CLAIM = "efficacy_claim"    # 功效宣称
    COMPETITOR_LOGO = "competitor_logo"  # 竞品露出
    DURATION_SHORT = "duration_short"    # 时长不足
    MENTION_MISSING = "mention_missing"  # 品牌提及不足
    BRAND_SAFETY = "brand_safety"        # 品牌安全风险


class ViolationSource(str, Enum):
    """违规来源"""
    TEXT = "text"        # 文本/脚本
    SPEECH = "speech"    # 语音（ASR）
    SUBTITLE = "subtitle"  # 字幕（OCR）
    VISUAL = "visual"    # 画面（CV）


class SoftRiskAction(str, Enum):
    """软性风控动作"""
    CONFIRM = "confirm"  # 需要二次确认
    NOTE = "note"        # 需要填写备注


class SoftRiskWarning(BaseModel):
    """软性风控提示（Warn-only）"""
    code: str = Field(..., description="提示类型代码")
    message: str = Field(..., description="提示内容")
    action_required: SoftRiskAction = Field(..., description="要求动作")
    blocking: bool = Field(default=False, description="是否阻断（默认不阻断）")
    context: Optional[dict] = Field(None, description="附加上下文")


class SoftRiskContext(BaseModel):
    """软性风控输入上下文"""
    violation_rate: Optional[float] = Field(None, ge=0, le=1, description="违规率")
    violation_threshold: Optional[float] = Field(None, ge=0, le=1, description="违规率阈值")
    asr_confidence: Optional[float] = Field(None, ge=0, le=1, description="ASR 置信度")
    ocr_confidence: Optional[float] = Field(None, ge=0, le=1, description="OCR 置信度")
    has_history_violation: Optional[bool] = Field(None, description="是否有历史类似违规")


# ==================== 通用模型 ====================

class Position(BaseModel):
    """文本位置"""
    start: int = Field(..., description="起始位置")
    end: int = Field(..., description="结束位置")


class Violation(BaseModel):
    """违规项（统一结构）"""
    type: ViolationType = Field(..., description="违规类型")
    content: str = Field(..., description="违规内容")
    severity: RiskLevel = Field(..., description="严重程度")
    suggestion: str = Field(..., description="修改建议")

    # 文本审核字段
    position: Optional[Position] = Field(None, description="文本位置（脚本审核）")

    # 视频审核字段
    timestamp: Optional[float] = Field(None, description="开始时间戳（秒）")
    timestamp_end: Optional[float] = Field(None, description="结束时间戳（秒）")
    source: Optional[ViolationSource] = Field(None, description="违规来源（视频审核）")


# ==================== 脚本预审 ====================

class ScriptReviewRequest(BaseModel):
    """脚本预审请求"""
    content: str = Field(..., min_length=1, description="脚本内容")
    platform: Platform = Field(..., description="投放平台")
    brand_id: str = Field(..., description="品牌 ID")
    required_points: Optional[list[str]] = Field(None, description="必要卖点列表")
    blacklist_words: Optional[list[dict]] = Field(None, description="Brief 黑名单词 [{word, reason}]")
    soft_risk_context: Optional[SoftRiskContext] = Field(None, description="软性风控上下文")


class ScriptReviewResponse(BaseModel):
    """
    脚本预审响应

    结构：
    - score: 合规分数 0-100
    - summary: 整体摘要
    - violations: 违规项列表，每项包含 suggestion
    - missing_points: 遗漏的卖点（可选）
    """
    score: int = Field(..., ge=0, le=100, description="合规分数")
    summary: str = Field(..., description="审核摘要")
    violations: list[Violation] = Field(default_factory=list, description="违规项列表")
    missing_points: Optional[list[str]] = Field(None, description="遗漏的卖点")
    soft_warnings: list[SoftRiskWarning] = Field(default_factory=list, description="软性风控提示")


# ==================== 视频审核 ====================

class VideoReviewRequest(BaseModel):
    """视频审核请求"""
    video_url: HttpUrl = Field(..., description="视频 URL")
    platform: Platform = Field(..., description="投放平台")
    brand_id: str = Field(..., description="品牌 ID")
    creator_id: str = Field(..., description="达人 ID")
    competitors: Optional[list[str]] = Field(None, description="竞品列表")
    requirements: Optional[dict] = Field(None, description="审核要求（时长、频次等）")


class VideoReviewSubmitResponse(BaseModel):
    """视频审核提交响应（202 Accepted）"""
    review_id: str = Field(..., description="审核任务 ID")
    status: TaskStatus = Field(default=TaskStatus.PENDING, description="任务状态")


class VideoReviewProgressResponse(BaseModel):
    """视频审核进度响应"""
    review_id: str = Field(..., description="审核任务 ID")
    status: TaskStatus = Field(..., description="任务状态")
    progress: int = Field(..., ge=0, le=100, description="进度百分比")
    current_step: str = Field(..., description="当前处理步骤")


class VideoReviewResultResponse(BaseModel):
    """
    视频审核结果响应（200 OK）

    结构与脚本审核一致：
    - score: 合规分数
    - summary: 整体摘要
    - violations: 违规项列表，每项包含 timestamp 和 suggestion
    """
    review_id: str = Field(..., description="审核任务 ID")
    status: TaskStatus = Field(default=TaskStatus.COMPLETED, description="任务状态")
    score: int = Field(..., ge=0, le=100, description="合规分数")
    summary: str = Field(..., description="审核摘要")
    violations: list[Violation] = Field(default_factory=list, description="违规项列表")
    soft_warnings: list[SoftRiskWarning] = Field(default_factory=list, description="软性风控提示")