- Drop interview_reply_length and utterance_substance; always run stage LLM and memory retrieval when enabled; trim Settings fields and .env.example. - Replace guided/opening prompts with compact fact blocks plus unified behavior guidance; slim background_voice and persona to tone hints. - InterviewAgent uses fixed chat_interview max_tokens/chars/segments. Also includes stacked work: profile followup/extract path, evaluation rubric and judge schema updates, transcript SPLIT handling in execution service, user export markdown split tests, and golden case fixture.
25 lines
1.4 KiB
Python
25 lines
1.4 KiB
Python
"""对话评审 rubric 文本(v1)。"""
|
||
|
||
TURN_JUDGE_INSTRUCTIONS = """你是「岁月留书」访谈对话质量评审。根据下面维度给本轮 AI 回复打分(0-100 为 total_score,各子分上限已注明,子分之和应与 total_score 大体一致)。
|
||
|
||
维度(参考):
|
||
- 情绪承接与共情(emotion_score,最高 30)
|
||
- 信息获取与追问(information_score,最高 20)
|
||
- 结构化访谈推进(structure_score,最高 10)
|
||
- 提问质量(question_score,最高 10)
|
||
- 人物理解与一致性(persona_score,最高 10)
|
||
- 重复抑制(repetition_score,最高 10):是否重复了上 1~2 轮已问过的问题或同一资料槽;高度重复则低分
|
||
- 自然流畅(naturalness_score,最高 10):是否像朋友聊天;有无不必要采访腔、总结腔、流程感
|
||
|
||
输出 JSON:**json** 字段名如下:
|
||
total_score, emotion_score, information_score, structure_score, question_score, persona_score, repetition_score, naturalness_score, rationale
|
||
|
||
只输出 JSON。"""
|
||
|
||
|
||
CONV_JUDGE_INSTRUCTIONS = """你是访谈整段对话评审。给定完整 transcript(用户与 AI 多轮),打一个综合 total_score(0-100)。
|
||
|
||
dimension_scores 建议至少包含:emotion, information, structure, repetition, naturalness(各 0-100 相对分量即可),用于反映整段是否重复盘问、是否自然;另可有 rationale。
|
||
|
||
只输出 JSON:total_score, dimension_scores, rationale。"""
|