wsk/deerflow-code
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

样式修改

Browse Source
This commit is contained in:
董宏凯 committed 2026-05-21 17:22:40 +08:00
1 parent ca13a9199f
commit 065bd8f28a
2 files changed
+112 -38
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py
+62 -29
View File
@@ -19,8 +19,8 @@ from ._utils import llm_json, safe_get_stream_writer
logger = logging.getLogger(__name__)
# 素材总量上限边界：config.yaml 的 ai_writing.max_materials，非法值回退默认 20
_DEFAULT_MAX_MATERIALS = 20
# 素材总量上限边界：config.yaml 的 ai_writing.max_materials，非法值回退默认 50
_DEFAULT_MAX_MATERIALS = 50
_MIN_MAX_MATERIALS = 1
_MAX_MAX_MATERIALS = 100
@@ -93,12 +93,18 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
article_type = state.get("article_type", "analysis")
target_audience = state.get("target_audience", "普通读者")
# 检查是否是补充搜索（PAUSE-1 re_search action）
# 检查干预模式（PAUSE-1 素材确认 / 素材不足求助）
intervention = state.get("last_user_intervention") or {}
extra_keywords: list[str] = intervention.get("_extra_keywords") or intervention.get("extra_keywords") or []
action = intervention.get("action")
# 素材不足求助点触发的「补充检索」：保留原大纲，检索完直接回写作、不走素材确认/大纲
is_supplement = intervention.get("action") == "supplement_search"
is_supplement = action == "supplement_search"
# PAUSE-1 素材确认卡用户点「追加检索」并填了新关键词：
# - 跳过 LLM 重新生成检索词（节省一次调用 + 尊重用户意图）
# - 只检索 extra_keywords 这几个新词，不重检索之前 LLM 出过的词
# - 保留 existing material_package，按 URL 去重合并
is_append_only = action == "re_search" and bool(extra_keywords)
# 用户指定的检索关键词数量（1~10），决定生成多少个检索词、检索多少个方向
keyword_count = _resolve_keyword_count(state.get("keyword_count"))
@@ -119,32 +125,43 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
writer({"type": "research_progress", "message": message})
# Step 1: 提取关键词
_emit("正在生成检索词…")
kw_result = await llm_json(
EXTRACT_KEYWORDS_SYSTEM,
EXTRACT_KEYWORDS_USER.format(
user_intent=intent,
article_type=article_type,
target_audience=target_audience,
keyword_count=keyword_count,
),
fallback={"keywords": [intent]},
config=config,
model_name=keyword_model,
)
# 清洗关键词：剔除占位符 / 空值，全空时回退写作意图
keywords: list[str] = _clean_keywords(kw_result.get("keywords"), intent)
if extra_keywords:
keywords = list(dict.fromkeys(keywords + extra_keywords))
kw_result: dict = {} # 走追加分支时跳过 LLM 调用，但下方 notice 拼接需要兜底值
if is_append_only:
# 追加模式：直接用用户填的关键词，不再调 LLM
keywords: list[str] = _clean_keywords(extra_keywords, intent)
_emit(f"追加检索 {len(keywords)} 个关键词…")
else:
_emit("正在生成检索词…")
kw_result = await llm_json(
EXTRACT_KEYWORDS_SYSTEM,
EXTRACT_KEYWORDS_USER.format(
user_intent=intent,
article_type=article_type,
target_audience=target_audience,
keyword_count=keyword_count,
),
fallback={"keywords": [intent]},
config=config,
model_name=keyword_model,
)
# 清洗关键词：剔除占位符 / 空值，全空时回退写作意图
keywords = _clean_keywords(kw_result.get("keywords"), intent)
if extra_keywords:
keywords = list(dict.fromkeys(keywords + extra_keywords))
# Step 2: 并发搜索
provider = get_search_provider(ai_writing_config)
# 按用户指定的关键词数量截断；补充搜索时不截断用户额外添加的关键词
search_limit = keyword_count
if extra_keywords:
search_limit = max(search_limit, len(keywords))
search_keywords = keywords[:search_limit]
# 按用户指定的关键词数量截断。
# 追加模式：keywords 本身就是用户填的，全部检索不截断（用户填 N 个就搜 N 个）。
# 其他模式：受 keyword_count 限制，但 extra_keywords 在内时上限放宽到 keywords 总数。
if is_append_only:
search_keywords = keywords
else:
search_limit = keyword_count
if extra_keywords:
search_limit = max(search_limit, len(keywords))
search_keywords = keywords[:search_limit]
total = len(search_keywords)
# 每个检索词分到的检索条数 = 总量上限 / 检索词数（向上取整）
per_keyword = _per_keyword_limit(max_materials, total)
@@ -168,10 +185,14 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
all_results_nested: list[list[SearchResult]] = await asyncio.gather(*search_tasks, return_exceptions=True)
# 合并去重
# - 首次检索 / 普通 re_search（不带 extra_keywords，用户只想 LLM 重出）：旧素材丢弃
# - 素材不足求助 supplement_search：保留旧素材 + 追加新检索结果
# - PAUSE-1 追加检索 is_append_only：保留旧素材 + 仅检索新关键词
seen_urls: set[str] = set()
raw_materials: list[Material] = []
existing_pkg = state.get("material_package")
if existing_pkg and intervention.get("action") != "re_search":
preserve_existing = existing_pkg and (is_supplement or is_append_only)
if preserve_existing:
for m in existing_pkg.get("materials", []):
seen_urls.add(m.get("url", ""))
raw_materials.append(m)
@@ -235,7 +256,13 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
if not materials:
summary = "未检索到相关素材，可重新搜索或在无素材的情况下继续写作"
pkg = MaterialPackage(keywords=keywords, materials=materials, summary=summary)
# 追加模式下，pkg.keywords 体现到目前为止累积的检索词（旧 + 新追加），
# 让前端列表 / 历史回显能看到完整的搜索路径。
pkg_keywords = keywords
if is_append_only and existing_pkg:
old_kws = list(existing_pkg.get("keywords") or [])
pkg_keywords = list(dict.fromkeys(old_kws + keywords))
pkg = MaterialPackage(keywords=pkg_keywords, materials=materials, summary=summary)
# 容错告警：检索接口 / 模型调用出错时以非致命 notice 提示用户，不打断写作流程
notices: list[dict] = []
@@ -253,12 +280,18 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
raw="\n\n----------\n\n".join(raw_dumps[:3]),
))
# 进度事件文案：追加模式下让用户看到只搜了哪几个新词
step_start_msg = (
f"追加检索关键词：{', '.join(keywords[:3])}"
if is_append_only
else f"搜索关键词：{', '.join(keywords[:3])}"
)
return {
"material_package": pkg,
# 补充检索：直接回写作（沿用原大纲）；常规检索：进入素材确认暂停点
"status": "writing_draft" if is_supplement else "awaiting_material_confirm",
"progress_events": [
{"type": "step_start", "agent_name": "素材收集专家", "message": f"搜索关键词：{', '.join(keywords[:3])}"},
{"type": "step_start", "agent_name": "素材收集专家", "message": step_start_msg},
*notices,
{"type": "materials_ready", "agent_name": "素材收集专家", "message": summary},
],