Fix/bug en zh (#382)
* [fix]The log retains genuine alerts and errors, while filtering out unnecessary noise. * [fix]Scenario English and Chinese, emotion specifications * [fix]Change the "no data" scenario from 0.0 to None
This commit is contained in:
乐力齐
@@ -349,19 +349,39 @@ async def render_emotion_suggestions_prompt(
|
||||
import json
|
||||
|
||||
# 预处理 emotion_distribution 为 JSON 字符串
|
||||
# 如果是中文，将 emotion_distribution 的 key 翻译为中文
|
||||
emotion_distribution = health_data.get('emotion_distribution', {})
|
||||
if language == "zh":
|
||||
emotion_type_zh = {
|
||||
'joy': '喜悦', 'sadness': '悲伤', 'anger': '愤怒',
|
||||
'fear': '恐惧', 'surprise': '惊讶', 'neutral': '中性'
|
||||
}
|
||||
emotion_distribution = {
|
||||
emotion_type_zh.get(k, k): v for k, v in emotion_distribution.items()
|
||||
}
|
||||
emotion_distribution_json = json.dumps(
|
||||
health_data.get('emotion_distribution', {}),
|
||||
emotion_distribution,
|
||||
ensure_ascii=False,
|
||||
indent=2
|
||||
)
|
||||
|
||||
# 翻译 dominant_negative_emotion
|
||||
dominant_negative_translated = None
|
||||
dominant_neg = patterns.get('dominant_negative_emotion')
|
||||
if dominant_neg and language == "zh":
|
||||
emotion_type_zh_map = {
|
||||
'sadness': '悲伤', 'anger': '愤怒', 'fear': '恐惧'
|
||||
}
|
||||
dominant_negative_translated = emotion_type_zh_map.get(dominant_neg, dominant_neg)
|
||||
|
||||
template = prompt_env.get_template("generate_emotion_suggestions.jinja2")
|
||||
rendered_prompt = template.render(
|
||||
health_data=health_data,
|
||||
patterns=patterns,
|
||||
user_profile=user_profile,
|
||||
emotion_distribution_json=emotion_distribution_json,
|
||||
language=language
|
||||
language=language,
|
||||
dominant_negative_translated=dominant_negative_translated
|
||||
)
|
||||
|
||||
# 记录渲染结果到提示日志
|
||||
|
||||
@@ -1,10 +1,23 @@
|
||||
{% if language == "en" %}
|
||||
You are a professional mental health consultant. Based on the following user's emotional health data and personal information, generate 3-5 personalized emotional improvement suggestions.
|
||||
|
||||
## Core Principle (Highest Priority)
|
||||
|
||||
**You must strictly base your suggestions on the emotion distribution data provided below. As long as any emotion type has a count ≥ 1, that emotion EXISTS and you must acknowledge and address it in your suggestions. You must NEVER claim an emotion is "zero" or "absent" when its count is ≥ 1.**
|
||||
|
||||
Specific rules:
|
||||
1. Carefully check the count for each emotion type in "Emotion Distribution" — count ≥ 1 means the emotion exists
|
||||
2. Even if an emotion appeared only once, you must mention it in health_summary or suggestions and provide targeted advice
|
||||
3. Never state that an emotion is "zero" or "non-existent" unless its count in the distribution data is truly 0
|
||||
4. If positive emotions (e.g., Joy) exist, health_summary must affirm this positive signal
|
||||
5. If negative emotions (e.g., Sadness, Anger, Fear) exist even once, you must provide targeted improvement suggestions
|
||||
6. A high proportion of neutral emotions does NOT mean other emotions are absent — address all non-zero emotions
|
||||
|
||||
## User Emotional Health Data
|
||||
|
||||
Health Score: {{ health_data.health_score }}/100
|
||||
Health Level: {{ health_data.level }}
|
||||
Total Emotion Records: {{ health_data.dimensions.positivity_rate.positive_count + health_data.dimensions.positivity_rate.negative_count + health_data.dimensions.positivity_rate.neutral_count }}
|
||||
|
||||
Dimension Analysis:
|
||||
- Positivity Rate: {{ health_data.dimensions.positivity_rate.score }}/100
|
||||
@@ -18,7 +31,7 @@ Dimension Analysis:
|
||||
- Resilience: {{ health_data.dimensions.resilience.score }}/100
|
||||
- Recovery Rate: {{ health_data.dimensions.resilience.recovery_rate }}
|
||||
|
||||
Emotion Distribution:
|
||||
Emotion Distribution (check each item — every emotion with count ≥ 1 must be reflected in suggestions):
|
||||
{{ emotion_distribution_json }}
|
||||
|
||||
## Emotion Pattern Analysis
|
||||
@@ -41,6 +54,7 @@ Please generate 3-5 personalized suggestions, each containing:
|
||||
5. actionable_steps: 3 specific executable steps
|
||||
|
||||
Also provide a health_summary (no more than 50 words) summarizing the user's overall emotional state.
|
||||
**The health_summary must truthfully reflect ALL non-zero emotions from the distribution data. Do not omit any emotion type that has appeared.**
|
||||
|
||||
Please return in JSON format as follows:
|
||||
{
|
||||
@@ -57,6 +71,7 @@ Please return in JSON format as follows:
|
||||
}
|
||||
|
||||
Notes:
|
||||
- CRITICAL: Any emotion with count ≥ 1 in the distribution MUST be acknowledged and addressed — never ignore or claim it is zero
|
||||
- Suggestions should be specific and actionable, avoid vague advice
|
||||
- Provide personalized suggestions based on user's interests and hobbies
|
||||
- Provide targeted suggestions for main issues (such as dominant negative emotions)
|
||||
@@ -66,10 +81,23 @@ Notes:
|
||||
{% else %}
|
||||
你是一位专业的心理健康顾问。请根据以下用户的情绪健康数据和个人信息，生成3-5条个性化的情绪改善建议。
|
||||
|
||||
## 核心原则（最高优先级）
|
||||
|
||||
**你必须严格基于下方提供的情绪分布数据来生成建议。只要某种情绪的出现次数 ≥ 1，就代表该情绪确实存在，你必须在建议中承认并回应这一情绪，绝对不能说"该情绪为零"或"没有该情绪"。**
|
||||
|
||||
具体规则：
|
||||
1. 仔细查看"情绪分布"中每种情绪的出现次数，次数 ≥ 1 即表示该情绪存在
|
||||
2. 即使某种情绪只出现了1次，也必须在 health_summary 或建议中提及并给出针对性建议
|
||||
3. 严禁在输出中声称某种情绪"为零"或"不存在"，除非该情绪在分布数据中确实为0次
|
||||
4. 如果正面情绪（如喜悦）存在，health_summary 中必须肯定这一积极信号
|
||||
5. 如果负面情绪（如悲伤、愤怒、恐惧）存在，即使只有1次，也必须给出针对性的改善建议
|
||||
6. 中性情绪占比高不代表没有其他情绪，必须同时关注所有非零情绪
|
||||
|
||||
## 用户情绪健康数据
|
||||
|
||||
健康分数：{{ health_data.health_score }}/100
|
||||
健康等级：{{ health_data.level }}
|
||||
情绪记录总数：{{ health_data.dimensions.positivity_rate.positive_count + health_data.dimensions.positivity_rate.negative_count + health_data.dimensions.positivity_rate.neutral_count }}条
|
||||
|
||||
维度分析：
|
||||
- 积极率：{{ health_data.dimensions.positivity_rate.score }}/100
|
||||
@@ -83,12 +111,12 @@ Notes:
|
||||
- 恢复力：{{ health_data.dimensions.resilience.score }}/100
|
||||
- 恢复率：{{ health_data.dimensions.resilience.recovery_rate }}
|
||||
|
||||
情绪分布：
|
||||
情绪分布（请逐项检查，次数≥1的情绪都必须在建议中体现）：
|
||||
{{ emotion_distribution_json }}
|
||||
|
||||
## 情绪模式分析
|
||||
|
||||
主要负面情绪：{{ patterns.dominant_negative_emotion|default('无') }}
|
||||
主要负面情绪：{{ dominant_negative_translated|default(patterns.dominant_negative_emotion)|default('无') }}
|
||||
情绪波动性：{{ patterns.emotion_volatility|default('未知') }}
|
||||
高强度情绪次数：{{ patterns.high_intensity_emotions|default([])|length }}
|
||||
|
||||
@@ -106,6 +134,7 @@ Notes:
|
||||
5. actionable_steps: 3个可执行的具体步骤
|
||||
|
||||
同时提供一个health_summary（不超过50字），概括用户的整体情绪状态。
|
||||
**health_summary 必须如实反映情绪分布中所有非零情绪的存在，不得遗漏任何已出现的情绪类型。**
|
||||
|
||||
请以JSON格式返回，格式如下：
|
||||
{
|
||||
@@ -122,9 +151,11 @@ Notes:
|
||||
}
|
||||
|
||||
注意事项：
|
||||
- 所有输出内容必须完全使用中文，严禁出现任何英文单词或短语（包括情绪类型名称如fear、sadness、anger等，必须使用对应的中文：恐惧、悲伤、愤怒等）
|
||||
- 再次强调：情绪分布中出现次数≥1的情绪必须在建议中被提及和回应，绝不能忽略或声称为零
|
||||
- 建议要具体、可执行，避免空泛
|
||||
- 结合用户的兴趣爱好提供个性化建议
|
||||
- 针对主要问题（如主要负面情绪）提供针对性建议
|
||||
- 优先级要合理分配（至少1个high，1-2个medium，其余low）
|
||||
- 优先级要合理分配（至少1个高，1-2个中，其余低）
|
||||
- 每个建议的3个步骤要循序渐进、易于实施
|
||||
{% endif %}
|
||||
|
||||
Reference in New Issue
Block a user