优化斗鱼粉丝日报定时任务与弹幕素材

1. 将斗鱼粉丝日报接入插件定时任务体系，支持后台独立启停与调度执行\n2. 为粉丝日报新增按群去重发送标记，避免任务补偿或多群订阅时重复推送\n3. 扩充日报传给LLM的弹幕材料，补充顺时序现场样本与场次故事线，提升语境完整度\n4. 提升斗鱼日报缓存版本，确保新链路生成结果不复用旧缓存
2026-04-29 13:19:19 +08:00
parent 3accd84bd1
commit 6d33081e03
2 changed files with 236 additions and 19 deletions
@@ -213,6 +213,18 @@ class DouyuDanmuSummaryHelper:
            "peak_buckets": cls._simplify_peak_buckets(peak_buckets),
            "representative_messages": cls._pick_representative_messages(organized_messages, bucket_stats),
            "raw_window_samples": cls._build_raw_window_samples(peak_buckets, per_bucket_limit=8),
+            # 给日报类 LLM 再补一层“按时间推进的现场切片”。
+            # 这样模型除了看热点窗口，还能顺着时间线理解气氛如何起、如何变、最后怎么收，
+            # 对粉丝日报这类强调“节目效果”和“接梗链路”的文本尤其有帮助。
+            "chronological_samples": cls._build_chronological_samples(organized_messages, limit=20),
+            # 每个 session 单独给一个轻量摘要，避免多场直播合并后，
+            # 模型只看到全局热点而丢失“第一场在聊什么、第二场为什么突然转节奏”的信息。
+            "session_storyline": cls._build_session_storyline(
+                organized_messages,
+                bucket_stats,
+                top_terms_limit=8,
+                sample_limit=10,
+            ),
        }

    @staticmethod
@@ -834,6 +846,83 @@ class DouyuDanmuSummaryHelper:
            })
        return windows

+    @classmethod
+    def _build_chronological_samples(
+        cls,
+        messages: List[Dict[str, Any]],
+        limit: int = 20,
+    ) -> List[Dict[str, str]]:
+        """
+        从整场弹幕里按时间均匀抽取样本。
+        设计目的：
+        1. 热点窗口只能解释“最炸的几分钟”，但日报还需要理解整体节奏；
+        2. 顺时序样本能帮助 LLM 看到开场铺垫、中段起哄、尾段收束；
+        3. 对粉丝日报来说，这比单纯词频更容易还原“今天到底经历了什么”。
+        """
+        if not messages:
+            return []
+
+        indexes = {
+            int(round((len(messages) - 1) * idx / max(limit - 1, 1)))
+            for idx in range(min(limit, len(messages)))
+        }
+        selected: List[Dict[str, str]] = []
+        seen = set()
+        for idx in sorted(indexes):
+            item = messages[idx]
+            content = str(item.get("content") or "").strip()
+            if not content:
+                continue
+            normalized = cls._normalize_template_text(content)
+            if normalized and normalized in seen:
+                continue
+            if normalized:
+                seen.add(normalized)
+            selected.append({
+                "time": str(item.get("timestamp_text") or ""),
+                "nickname": str(item.get("nickname") or ""),
+                "content": content[:90],
+            })
+            if len(selected) >= limit:
+                break
+        return selected
+
+    @classmethod
+    def _build_session_storyline(
+        cls,
+        messages: List[Dict[str, Any]],
+        bucket_stats: List[Dict[str, Any]],
+        *,
+        top_terms_limit: int = 8,
+        sample_limit: int = 10,
+    ) -> Dict[str, Any]:
+        """
+        组装单场直播的轻量叙事骨架。
+        这里不追求大而全，而是给模型一个“这场直播从头到尾在发生什么”的概览，
+        让它在写日报时更容易把梗、情绪和时间顺序串起来。
+        """
+        first_message = messages[0] if messages else {}
+        last_message = messages[-1] if messages else {}
+        hottest_bucket = max(
+            bucket_stats,
+            key=lambda item: int(item.get("message_count", 0) or 0),
+            default={},
+        )
+        return {
+            "start_time": str(first_message.get("timestamp_text") or ""),
+            "end_time": str(last_message.get("timestamp_text") or ""),
+            "top_terms": cls._extract_top_terms(messages, limit=top_terms_limit),
+            "burst_terms": cls._build_burst_terms(messages)[:6],
+            "chronological_samples": cls._build_chronological_samples(messages, limit=sample_limit),
+            "hottest_moment": {
+                "start_time": str(hottest_bucket.get("start_time") or ""),
+                "message_count": int(hottest_bucket.get("message_count", 0) or 0),
+                "user_count": int(hottest_bucket.get("user_count", 0) or 0),
+                "top_terms": hottest_bucket.get("top_terms", [])[:6],
+                "sample_messages": hottest_bucket.get("sample_messages", [])[:6],
+            },
+        }
+
    @staticmethod
    def _simplify_peak_buckets(buckets: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
        simplified = []