136 lines
4.7 KiB
Python
136 lines
4.7 KiB
Python
"""
|
||
文字询价字段抽取入口。
|
||
|
||
本文件职责：生产走 DeepSeek B；单测可注入 facts；无网时只认「键：值」结构化行，不猜港口。
|
||
禁止：用正则把「上海到洛杉矶」拆成起运目的；禁止改六态。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import logging
|
||
import re
|
||
from typing import Any, Optional
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
_MODE_WORDS = (
|
||
("空运", "AIR"),
|
||
("海运", "SEA"),
|
||
("陆运", "LAND"),
|
||
)
|
||
|
||
# 仅「键：值」一行一条，键必须是合同字段或展示名
|
||
_LABEL_LINE = re.compile(
|
||
r"^(起运地|起运港|目的地|目的港|品名|件数|毛重|体积|重量|重量\(KG\)|体积\(CBM\)|"
|
||
r"包装方式|包装类型|货量|货物数量|整柜或拼柜|箱型箱量|报价日期|车型数量|"
|
||
r"贸易条款|货好时间|HS编码|HS|是否含油|是否含电|是否含磁)\s*[:：]\s*(.+)$"
|
||
)
|
||
_LABEL_INLINE = re.compile(
|
||
r"(起运地|起运港|目的地|目的港|品名|件数|毛重|体积|重量|重量\(KG\)|体积\(CBM\)|"
|
||
r"包装方式|包装类型|货量|货物数量|整柜或拼柜|箱型箱量|报价日期|车型数量|"
|
||
r"贸易条款|货好时间|HS编码|HS|是否含油|是否含电|是否含磁)\s*[:：]\s*(\S+)"
|
||
)
|
||
|
||
|
||
def detect_transport_mode(text: str) -> str:
|
||
"""
|
||
运输方式：空运/海运/陆运，或空运口语「飞」（南京飞吉隆坡）。
|
||
|
||
写了海运/陆运以书面为准。只说飞就是空运，补问件数时不得再问运输方式。
|
||
"""
|
||
raw = text or ""
|
||
found = [code for word, code in _MODE_WORDS if word in raw]
|
||
if len(found) == 1:
|
||
return found[0]
|
||
if len(found) > 1:
|
||
return ""
|
||
if "飞" in raw:
|
||
return "AIR"
|
||
return ""
|
||
|
||
|
||
def parse_labeled_facts(text: str) -> dict[str, str]:
|
||
"""只采集「字段名：值」，整行优先；同一句里的标签也收。不猜港口。"""
|
||
facts: dict[str, str] = {}
|
||
raw = text or ""
|
||
for line in raw.splitlines():
|
||
m = _LABEL_LINE.match(line.strip())
|
||
if m:
|
||
facts[m.group(1)] = m.group(2).strip()
|
||
if facts:
|
||
return facts
|
||
for m in _LABEL_INLINE.finditer(raw):
|
||
facts[m.group(1)] = m.group(2).strip()
|
||
return facts
|
||
|
||
|
||
def extract_inquiry_snapshot(
|
||
text: str,
|
||
*,
|
||
injected_facts: Optional[dict[str, Any]] = None,
|
||
injected_mode: str = "",
|
||
b_result: Optional[dict[str, Any]] = None,
|
||
allow_b: bool = False,
|
||
chat_fn: Optional[Any] = None,
|
||
) -> dict[str, Any]:
|
||
"""
|
||
抽出业务线 + 字段快照。
|
||
|
||
优先 injected_*（单测）；其次合并标签行与 DeepSeek B（口语）。
|
||
标签行覆盖 B 的同名字段。B 未出现的字段保持空，不本地猜港口。
|
||
件数/毛重/体积若原话已带 TMS 单位（件/KGS/CBM），空字段可机读补上。
|
||
"""
|
||
from agent.schema.field_validate import harvest_oral_measures, normalize_facts
|
||
|
||
if injected_facts is not None:
|
||
mode = (injected_mode or detect_transport_mode(text) or "").upper()
|
||
return {
|
||
"business_line": mode,
|
||
"land_subtype": "",
|
||
"facts": harvest_oral_measures(text, normalize_facts(dict(injected_facts))),
|
||
"source": "injected",
|
||
}
|
||
mode = detect_transport_mode(text)
|
||
facts = parse_labeled_facts(text)
|
||
source = "labeled_or_mode"
|
||
b_payload = b_result
|
||
if b_payload is None and (allow_b or chat_fn is not None):
|
||
from agent.llm.mode_extract_fields import invoke_extract_fields
|
||
|
||
b_payload = invoke_extract_fields(
|
||
text, allow_network=allow_b, chat_fn=chat_fn
|
||
)
|
||
if b_payload and b_payload.get("ok"):
|
||
merged = dict(b_payload.get("facts") or {})
|
||
merged.update(facts)
|
||
facts = merged
|
||
mode = mode or str(b_payload.get("business_line") or "")
|
||
source = "deepseek_b"
|
||
facts = harvest_oral_measures(text, normalize_facts(facts))
|
||
if not mode and facts.get("起运港") and facts.get("目的港") and "飞" in (text or ""):
|
||
mode = "AIR"
|
||
logger.info("extract.snapshot mode=%s keys=%s src=%s", mode or "-", sorted(facts), source)
|
||
return {
|
||
"business_line": mode,
|
||
"land_subtype": (b_payload or {}).get("land_subtype") or "",
|
||
"facts": facts,
|
||
"source": source,
|
||
}
|
||
|
||
|
||
def extract_collab_fields(text: str) -> dict[str, str]:
|
||
"""
|
||
群里协同字段：只收带标签的非空值。
|
||
|
||
闲聊、报价金额、附件本轮不进这里。HS 折成 HS编码。
|
||
"""
|
||
from agent.schema.field_validate import SEA_COLLAB_FIELDS, normalize_facts
|
||
|
||
labeled = normalize_facts(parse_labeled_facts(text))
|
||
out: dict[str, str] = {}
|
||
for key in SEA_COLLAB_FIELDS:
|
||
val = (labeled.get(key) or "").strip()
|
||
if val:
|
||
out[key] = val
|
||
return out
|