324 lines
14 KiB
Python
324 lines
14 KiB
Python
import logging
|
||
import json
|
||
import os
|
||
from django.conf import settings
|
||
from openai import OpenAI
|
||
from .models import AIEvaluation
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
class BailianService:
|
||
def __init__(self):
|
||
self.api_key = getattr(settings, 'DASHSCOPE_API_KEY', None)
|
||
if not self.api_key:
|
||
self.api_key = os.environ.get("DASHSCOPE_API_KEY")
|
||
|
||
if self.api_key:
|
||
self.client = OpenAI(
|
||
api_key=self.api_key,
|
||
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1"
|
||
)
|
||
else:
|
||
self.client = None
|
||
logger.warning("DASHSCOPE_API_KEY not configured.")
|
||
|
||
def evaluate_task(self, evaluation: AIEvaluation):
|
||
"""
|
||
执行AI评估
|
||
"""
|
||
if not self.client:
|
||
evaluation.status = AIEvaluation.Status.FAILED
|
||
evaluation.error_message = "服务未配置 (DASHSCOPE_API_KEY missing)"
|
||
evaluation.save()
|
||
return
|
||
|
||
task = evaluation.task
|
||
if not task.transcription:
|
||
evaluation.status = AIEvaluation.Status.FAILED
|
||
evaluation.error_message = "关联任务无逐字稿内容"
|
||
evaluation.save()
|
||
return
|
||
|
||
evaluation.status = AIEvaluation.Status.PROCESSING
|
||
evaluation.save()
|
||
|
||
try:
|
||
prompt = evaluation.prompt
|
||
content = task.transcription
|
||
|
||
# 准备章节/时间戳数据以辅助分析发言节奏
|
||
chapter_context = ""
|
||
if task.auto_chapters_data:
|
||
try:
|
||
chapters_str = ""
|
||
# 处理特定的 AutoChapters 结构
|
||
# 格式: {"AutoChapters": [{"Id": 1, "Start": 740, "End": 203436, "Headline": "...", "Summary": "..."}, ...]}
|
||
if isinstance(task.auto_chapters_data, dict) and 'AutoChapters' in task.auto_chapters_data:
|
||
chapters = task.auto_chapters_data['AutoChapters']
|
||
if isinstance(chapters, list):
|
||
chapter_lines = []
|
||
for ch in chapters:
|
||
# 毫秒转 MM:SS
|
||
start_ms = ch.get('Start', 0)
|
||
end_ms = ch.get('End', 0)
|
||
start_str = f"{start_ms // 60000:02d}:{(start_ms // 1000) % 60:02d}"
|
||
end_str = f"{end_ms // 60000:02d}:{(end_ms // 1000) % 60:02d}"
|
||
|
||
headline = ch.get('Headline', '无标题')
|
||
summary = ch.get('Summary', '')
|
||
|
||
line = f"- [{start_str} - {end_str}] {headline}"
|
||
if summary:
|
||
line += f"\n 摘要: {summary}"
|
||
chapter_lines.append(line)
|
||
|
||
chapters_str = "\n".join(chapter_lines)
|
||
|
||
# 如果上面的解析为空（或者格式不匹配），回退到通用 JSON dump
|
||
if not chapters_str:
|
||
if isinstance(task.auto_chapters_data, (dict, list)):
|
||
chapters_str = json.dumps(task.auto_chapters_data, ensure_ascii=False, indent=2)
|
||
else:
|
||
chapters_str = str(task.auto_chapters_data)
|
||
|
||
chapter_context = f"\n\n【章节与时间戳信息】\n{chapters_str}\n\n(提示：请结合上述章节时间戳信息，分析发言者的语速、节奏变化及停顿情况。)"
|
||
except Exception as e:
|
||
logger.warning(f"Failed to process auto_chapters_data: {e}")
|
||
|
||
# 截断过长的内容以防止超出Token限制 (简单处理，取前10000字)
|
||
if len(content) > 10000:
|
||
content = content[:10000] + "...(内容过长已截断)"
|
||
|
||
# Construct messages
|
||
messages = [
|
||
{'role': 'system', 'content': 'You are a helpful assistant designed to output JSON.'},
|
||
{'role': 'user', 'content': f"{prompt}\n\n以下是需要评估的内容：\n{content}{chapter_context}"}
|
||
]
|
||
|
||
# 增加重试机制 (最多重试3次)
|
||
completion = None
|
||
last_error = None
|
||
import time
|
||
|
||
for attempt in range(3):
|
||
try:
|
||
completion = self.client.chat.completions.create(
|
||
model=evaluation.model_selection,
|
||
messages=messages,
|
||
response_format={"type": "json_object"}
|
||
)
|
||
break # 成功则跳出循环
|
||
except Exception as e:
|
||
last_error = e
|
||
logger.warning(f"AI Evaluation attempt {attempt+1}/3 failed for eval {evaluation.id}: {e}")
|
||
if attempt < 2:
|
||
time.sleep(2 * (attempt + 1)) # 简单的指数退避
|
||
|
||
if not completion:
|
||
raise last_error or Exception("AI Service call failed after retries")
|
||
|
||
response_content = completion.choices[0].message.content
|
||
# Convert to dict for storage
|
||
raw_response = completion.model_dump()
|
||
|
||
evaluation.raw_response = raw_response
|
||
|
||
# Parse JSON
|
||
try:
|
||
result = json.loads(response_content)
|
||
evaluation.score = result.get('score')
|
||
evaluation.evaluation = result.get('evaluation') or result.get('comment')
|
||
|
||
# 尝试获取推理过程（如果模型返回了）
|
||
evaluation.reasoning = result.get('reasoning') or result.get('analysis')
|
||
|
||
if not evaluation.reasoning:
|
||
# 如果JSON里没有，把整个JSON作为推理参考
|
||
evaluation.reasoning = json.dumps(result, ensure_ascii=False, indent=2)
|
||
|
||
evaluation.status = AIEvaluation.Status.COMPLETED
|
||
except json.JSONDecodeError:
|
||
evaluation.status = AIEvaluation.Status.FAILED
|
||
evaluation.error_message = f"无法解析JSON响应: {response_content}"
|
||
evaluation.reasoning = response_content
|
||
|
||
evaluation.save()
|
||
|
||
# 同步结果到参赛项目 (如果关联了)
|
||
self._sync_evaluation_to_project(evaluation)
|
||
|
||
return evaluation
|
||
|
||
except Exception as e:
|
||
logger.error(f"AI Evaluation failed: {e}")
|
||
evaluation.status = AIEvaluation.Status.FAILED
|
||
evaluation.error_message = str(e)
|
||
evaluation.save()
|
||
return evaluation
|
||
|
||
def _sync_evaluation_to_project(self, evaluation: AIEvaluation):
|
||
"""
|
||
将AI评估结果同步到关联的参赛项目（评分和评语）
|
||
"""
|
||
try:
|
||
task = evaluation.task
|
||
if not task.project:
|
||
return
|
||
|
||
project = task.project
|
||
competition = project.competition
|
||
|
||
# 1. 确定评委身份 (Based on Template)
|
||
# 用户要求：评委显示的是模板名称
|
||
template_name = evaluation.template.name if evaluation.template else "AI智能评委"
|
||
# 使用固定前缀 + template_id 确保唯一性，这样同一个模板在不同项目里是同一个评委
|
||
openid = f"ai_judge_{evaluation.template.id}" if evaluation.template else "ai_judge_default"
|
||
|
||
# 延迟导入以避免循环依赖
|
||
from shop.models import WeChatUser
|
||
from competition.models import CompetitionEnrollment, Score, Comment, ScoreDimension
|
||
|
||
# 获取或创建虚拟评委用户
|
||
user, created = WeChatUser.objects.get_or_create(
|
||
openid=openid,
|
||
defaults={
|
||
'nickname': template_name,
|
||
'avatar_url': 'https://ui-avatars.com/api/?name=AI&background=random&color=fff'
|
||
}
|
||
)
|
||
|
||
# 如果名字不匹配（比如模板改名了），更新它
|
||
if user.nickname != template_name:
|
||
user.nickname = template_name
|
||
user.save(update_fields=['nickname'])
|
||
|
||
# 2. 确保评委已报名 (Enrollment)
|
||
enrollment, _ = CompetitionEnrollment.objects.get_or_create(
|
||
competition=competition,
|
||
user=user,
|
||
defaults={
|
||
'role': 'judge',
|
||
'status': 'approved'
|
||
}
|
||
)
|
||
|
||
# 3. 同步评分 (Score)
|
||
if evaluation.score is not None:
|
||
# 尝试找到匹配的维度
|
||
dimensions = competition.score_dimensions.all()
|
||
target_dimension = None
|
||
|
||
# 0. 优先使用模板配置的维度
|
||
if evaluation.template and evaluation.template.score_dimension:
|
||
# 检查配置的维度是否属于当前比赛
|
||
if evaluation.template.score_dimension.competition_id == competition.id:
|
||
target_dimension = evaluation.template.score_dimension
|
||
else:
|
||
# 如果不属于当前比赛(跨比赛复用模板)，尝试查找同名维度
|
||
target_dimension = dimensions.filter(name=evaluation.template.score_dimension.name).first()
|
||
|
||
# 1. 如果未配置或未找到，尝试匹配 "AI Rating" (用户指定默认值)
|
||
if not target_dimension:
|
||
target_dimension = dimensions.filter(name__iexact="AI Rating").first()
|
||
|
||
# 2. 尝试匹配包含 "AI" 的维度
|
||
if not target_dimension:
|
||
for dim in dimensions:
|
||
if "AI" in dim.name.upper():
|
||
target_dimension = dim
|
||
break
|
||
|
||
# 3. 尝试匹配模板名称
|
||
if not target_dimension:
|
||
target_dimension = dimensions.filter(name=template_name).first()
|
||
|
||
# 4. 最后兜底：使用第一个维度
|
||
if not target_dimension and dimensions.exists():
|
||
target_dimension = dimensions.first()
|
||
|
||
if target_dimension:
|
||
Score.objects.update_or_create(
|
||
project=project,
|
||
judge=enrollment,
|
||
dimension=target_dimension,
|
||
defaults={'score': evaluation.score}
|
||
)
|
||
logger.info(f"Synced AI score {evaluation.score} to project {project.id} dimension {target_dimension.name}")
|
||
|
||
# 4. 同步评语 (Comment)
|
||
if evaluation.evaluation:
|
||
# 检查是否已存在该评委的评语，避免重复
|
||
comment = Comment.objects.filter(project=project, judge=enrollment).first()
|
||
if comment:
|
||
comment.content = evaluation.evaluation
|
||
comment.save()
|
||
else:
|
||
Comment.objects.create(
|
||
project=project,
|
||
judge=enrollment,
|
||
content=evaluation.evaluation
|
||
)
|
||
logger.info(f"Synced AI comment to project {project.id}")
|
||
|
||
except Exception as e:
|
||
logger.error(f"Failed to sync evaluation to project: {e}")
|
||
|
||
def summarize_task(self, task):
|
||
"""
|
||
对转写任务进行总结
|
||
"""
|
||
if not self.client:
|
||
logger.warning("BailianService not initialized, skipping summary.")
|
||
return
|
||
|
||
if not task.transcription:
|
||
logger.warning(f"Task {task.id} has no transcription, skipping summary.")
|
||
return
|
||
|
||
try:
|
||
content = task.transcription
|
||
# 简单截断防止过长
|
||
if len(content) > 15000:
|
||
content = content[:15000] + "...(内容过长已截断)"
|
||
|
||
# 准备上下文数据
|
||
context_data = ""
|
||
if task.summary_data:
|
||
context_data += f"\n\n【总结原始数据】\n{json.dumps(task.summary_data, ensure_ascii=False, indent=2)}"
|
||
|
||
if task.auto_chapters_data:
|
||
context_data += f"\n\n【章节原始数据】\n{json.dumps(task.auto_chapters_data, ensure_ascii=False, indent=2)}"
|
||
|
||
system_prompt = f"""你是一个专业的会议/内容总结助手。请根据提供的【转写文本】、【总结原始数据】和【章节原始数据】，生成一份结构清晰、内容详实的总结报告。
|
||
|
||
请按照以下结构输出（Markdown格式）：
|
||
1. **标题**：基于内容生成一个合适的标题。
|
||
2. **核心摘要**：简要概括主要内容。
|
||
3. **主要观点/话题**：结合思维导图数据，列出关键话题和层级。
|
||
4. **章节速览**：结合章节数据，列出时间点和主要内容。[HH:MM:SS]格式来把章节列出来
|
||
5. **问答精选**（如果有）：基于问答总结数据，列出重要问答。
|
||
|
||
请确保语言通顺，重点突出，能够还原内容的逻辑结构。"""
|
||
|
||
user_content = f"以下是需要总结的内容：\n\n【转写文本】\n{content}{context_data}"
|
||
|
||
messages = [
|
||
{'role': 'system', 'content': system_prompt},
|
||
{'role': 'user', 'content': user_content}
|
||
]
|
||
|
||
# 使用 qwen-plus 作为默认模型
|
||
completion = self.client.chat.completions.create(
|
||
model="qwen-plus",
|
||
messages=messages
|
||
)
|
||
|
||
summary_content = completion.choices[0].message.content
|
||
task.summary = summary_content
|
||
task.save(update_fields=['summary'])
|
||
|
||
logger.info(f"Task {task.id} summary generated successfully.")
|
||
|
||
except Exception as e:
|
||
logger.error(f"Failed to generate summary for task {task.id}: {e}")
|