吴话可说

调用 qwen3.7-max 对转写后的吴语台词进行批量翻译，同时补充文化典故注释。 由于戏剧台词口语化、押韵且含大量地方俗语，模型被提示保留戏剧语境， 输出严格的 JSON 数组以便直接写入数据库。

def translate_segments(segments: List[Segment]) -> None:
    batch_size = settings.translation_batch_size
    for i in range(0, len(segments), batch_size):
        batch = segments[i : i + batch_size]
        lines_text = "\n".join(
            f"{idx + 1}. {seg.speaker + '：' if seg.speaker else ''}{seg.wu_text}"
            for idx, seg in enumerate(batch)
        )
        prompt = f"""你是一位资深的吴语戏剧翻译。...
3. 判断每句台词的情感倾向，只能取 positive / neutral / negative 之一。
4. 输出严格的 JSON 数组... 每项字段：mandarin, english, notes, sentiment。
..."""
        content = _call_qwen([{"role": "user", "content": prompt}], max_tokens=8192)
        translations = _extract_json(content)
        for seg, trans in zip(batch, translations):
            seg.mandarin_text = str(trans.get("mandarin", "")).strip() or None
            seg.english_text = str(trans.get("english", "")).strip() or None
            seg.notes = str(trans.get("notes", "")).strip() or None
            raw_sentiment = str(trans.get("sentiment", "neutral")).strip().lower()
            seg.sentiment = raw_sentiment if raw_sentiment in {"positive", "neutral", "negative"} else "neutral"

在翻译阶段同步要求模型判断每句台词的情感倾向（positive / neutral / negative）， 将结果写入 Segment.sentiment 字段。该设计不增加额外模型调用， 同时满足大作业对「文本情感分析」关键技术的要求。

后端基于 Segment / Scene 数据计算发言者统计、情感分布、时间轴密度、场景情感走势； 前端使用 recharts 渲染为可交互图表，集成在作品详情页的「分析」Tab 中。

def compute_work_analytics(segments, scenes, duration_seconds):
    total_segments = len(segments)
    speaker_counter = Counter()
    speaker_seconds = defaultdict(float)
    for seg in segments:
        speaker = seg.speaker or "未知"
        speaker_counter[speaker] += 1
        speaker_seconds[speaker] += seg.end_seconds - seg.start_seconds

    sentiment_counter = Counter(_safe_sentiment(seg.sentiment) for seg in segments)

    minutes = max(1, math.ceil(duration_seconds / 60))
    timeline_density = [
        {"minute": m + 1, "count": sum(
            1 for seg in segments
            if seg.start_seconds < (m + 1) * 60 and seg.end_seconds > m * 60
        )}
        for m in range(minutes)
    ]
    return { ... }

<ResponsiveContainer width="100%" height="100%">
  <PieChart>
    <Pie data={sentimentData} dataKey="count" nameKey="name" ...>
      {sentimentData.map((entry, idx) => (
        <Cell key={idx} fill={entry.fill} />
      ))}
    </Pie>
    <Tooltip /><Legend />
  </PieChart>
</ResponsiveContainer>