ngfchl
163 lines
6.5 KiB
Python
163 lines
6.5 KiB
Python
"""知识库问答 handler"""
|
|
import asyncio
|
|
import html
|
|
import logging
|
|
import re
|
|
|
|
from telegram import Update
|
|
from telegram.ext import CommandHandler, ContextTypes, filters
|
|
|
|
import config
|
|
from services.message_utils import record_bot_message
|
|
|
|
logger = logging.getLogger("spam_guard")
|
|
|
|
|
|
def looks_like_markdown(text: str) -> bool:
|
|
return bool(re.search(r"(^|\n)(#{1,3}\s|[-*]\s|\d+\.\s|```)|\*\*[^*]+\*\*|\*[^*]+\*|`[^`]+`", text))
|
|
|
|
|
|
def optimize_plain_answer(text: str) -> str:
|
|
"""非 Markdown 回答也做一点可读性整理。"""
|
|
text = text.strip()
|
|
if looks_like_markdown(text):
|
|
return text
|
|
lines = [line.strip() for line in text.splitlines() if line.strip()]
|
|
if len(lines) <= 1:
|
|
return f"*回答*\n{text}"
|
|
return "*回答*\n" + "\n".join(f"- {line}" if not line.startswith(("-", "•", "✅", "⚠️", "📚")) else line for line in lines)
|
|
|
|
|
|
def markdown_to_html(text: str) -> str:
|
|
"""把常见 Markdown 转成 Telegram HTML，规避 Markdown 解析器对 _/链接/长文本的误判。"""
|
|
code_blocks: list[str] = []
|
|
|
|
def stash_code_block(match: re.Match) -> str:
|
|
code = html.escape(match.group(1).strip())
|
|
code_blocks.append(f"<pre><code>{code}</code></pre>")
|
|
return f"@@CODE_BLOCK_{len(code_blocks) - 1}@@"
|
|
|
|
text = re.sub(r"```(?:\w+)?\n?(.*?)```", stash_code_block, text, flags=re.S)
|
|
text = html.escape(text)
|
|
text = re.sub(r"^#{1,3}\s+(.+)$", r"<b>\1</b>", text, flags=re.M)
|
|
text = re.sub(r"\*\*([^*\n]+)\*\*", r"<b>\1</b>", text)
|
|
text = re.sub(r"(?<!\*)\*([^*\n]+)\*(?!\*)", r"<b>\1</b>", text)
|
|
text = re.sub(r"`([^`\n]+)`", r"<code>\1</code>", text)
|
|
for i, block in enumerate(code_blocks):
|
|
text = text.replace(f"@@CODE_BLOCK_{i}@@", block)
|
|
return text
|
|
|
|
|
|
def compact_answer(text: str, max_len: int = 3800) -> str:
|
|
"""尽量压到一条 Telegram 消息；超长时保留开头和关键尾部。"""
|
|
text = optimize_plain_answer(text)
|
|
if len(text) <= max_len:
|
|
return text
|
|
tail_markers = ["**文档**", "📚", "相关文档", "文档"]
|
|
tail = ""
|
|
for marker in tail_markers:
|
|
idx = text.rfind(marker)
|
|
if idx > max_len // 2:
|
|
tail = text[idx:].strip()
|
|
break
|
|
head_len = max_len - len(tail) - 80 if tail else max_len - 40
|
|
head = text[:max(1000, head_len)].rstrip()
|
|
return head + "\n\n...（内容较长已压缩，可追问具体步骤）" + (("\n\n" + tail) if tail else "")
|
|
|
|
|
|
def split_answer(text: str, max_len: int = 3900, max_parts: int = 2) -> list[str]:
|
|
"""优先单条；实在超长最多拆 2 条。"""
|
|
if len(text) <= max_len:
|
|
return [text]
|
|
parts = []
|
|
rest = text
|
|
while rest and len(parts) < max_parts:
|
|
cut = min(len(rest), max_len)
|
|
if cut < len(rest):
|
|
better = rest.rfind("\n\n", 0, cut)
|
|
if better > max_len // 2:
|
|
cut = better
|
|
parts.append(rest[:cut].strip())
|
|
rest = rest[cut:].strip()
|
|
if rest and parts:
|
|
parts[-1] += "\n\n...（后续内容已省略，可继续追问）"
|
|
return parts
|
|
|
|
|
|
async def safe_send(update: Update, context: ContextTypes.DEFAULT_TYPE, text: str, markdown: bool | None = None):
|
|
"""优先回复原消息；Markdown 内容转 HTML 发送，失败自动降级纯文本。"""
|
|
chat_id = update.effective_chat.id
|
|
use_markup = looks_like_markdown(text) if markdown is None else markdown
|
|
send_text = markdown_to_html(text) if use_markup else text
|
|
|
|
async def _record(msg, body):
|
|
await record_bot_message(msg, body)
|
|
return msg
|
|
|
|
if use_markup:
|
|
try:
|
|
return await _record(await update.message.reply_text(send_text, parse_mode="HTML", disable_web_page_preview=True), send_text)
|
|
except Exception as e:
|
|
if "Message to be replied not found" in str(e):
|
|
try:
|
|
return await _record(await context.bot.send_message(chat_id=chat_id, text=send_text, parse_mode="HTML", disable_web_page_preview=True), send_text)
|
|
except Exception:
|
|
return await _record(await context.bot.send_message(chat_id=chat_id, text=text, disable_web_page_preview=True), text)
|
|
logger.warning(f"HTML 格式发送失败，降级普通文本: {e}")
|
|
|
|
try:
|
|
return await _record(await update.message.reply_text(text, disable_web_page_preview=True), text)
|
|
except Exception as e:
|
|
if "Message to be replied not found" not in str(e):
|
|
logger.warning(f"发送回复失败，降级普通发送: {e}")
|
|
return await _record(await context.bot.send_message(chat_id=chat_id, text=text, disable_web_page_preview=True), text)
|
|
|
|
|
|
async def cmd_ask(update: Update, context: ContextTypes.DEFAULT_TYPE):
|
|
"""向知识库提问"""
|
|
query = " ".join(context.args) if context.args else ""
|
|
if not query:
|
|
reply = await safe_send(update, context,
|
|
"❓ 用法: /ask <你的问题>\n"
|
|
"例如: /ask 怎么安装收割机\n"
|
|
f"文档地址: {config.SITE_URL}"
|
|
)
|
|
asyncio.create_task(auto_delete(update.message, seconds=60))
|
|
asyncio.create_task(auto_delete(reply, seconds=60))
|
|
return
|
|
|
|
logger.debug(f"📩 用户提问: [{update.message.from_user.username}] {query[:80]}")
|
|
searching_msg = await safe_send(update, context, "🔍 正在搜索知识库并整理详细答案...")
|
|
asyncio.create_task(auto_delete(update.message, seconds=60))
|
|
|
|
try:
|
|
from services.knowledge_base import ask_knowledge_base
|
|
|
|
answer = compact_answer(await ask_knowledge_base(query))
|
|
replies = []
|
|
parts = split_answer(answer)
|
|
for i, part in enumerate(parts, 1):
|
|
prefix = f"*回答 {i}/{len(parts)}*\n\n" if len(parts) > 1 else ""
|
|
reply = await safe_send(update, context, prefix + part, markdown=looks_like_markdown(prefix + part))
|
|
replies.append(reply)
|
|
for reply in replies:
|
|
asyncio.create_task(auto_delete(reply, seconds=60))
|
|
asyncio.create_task(auto_delete(searching_msg, seconds=0))
|
|
except Exception as e:
|
|
logger.error(f"❌ 回答失败: {e}")
|
|
reply = await safe_send(update, context, f"❌ 回答失败: {e}")
|
|
asyncio.create_task(auto_delete(reply, seconds=60))
|
|
asyncio.create_task(auto_delete(searching_msg, seconds=0))
|
|
|
|
|
|
async def auto_delete(message, seconds=60):
|
|
try:
|
|
await asyncio.sleep(seconds)
|
|
await message.delete()
|
|
except Exception:
|
|
pass
|
|
|
|
|
|
def register_knowledge(app):
|
|
app.add_handler(CommandHandler("ask", cmd_ask, filters.ChatType.GROUPS))
|