添加情感分析prompt/类（暂未集成）

core/pipeline/emotion.py

@@ -0,0 +1,105 @@
+import re
+import json
+from enum import Enum
+
+from astrbot.api import logger
+from ..provider.llm.openai_source import ProviderOpenAI
+from ..util.prompts import EMOTION_ANALYSIS_PROMPT
+
+
+class EmotionTendency(Enum):
+    """情绪倾向枚举"""
+
+    POSITIVE = "positive"
+    """积极"""
+    NEGATIVE = "negative"
+    """消极"""
+    NEUTRAL = "neutral"
+    """中性"""
+
+
+class Emotion(Enum):
+    """情绪枚举"""
+
+    JOY = "joy"
+    """喜悦"""
+    CONTENTMENT = "contentment"
+    """满足"""
+    SURPRISE = "surprise"
+    """惊讶"""
+    NEUTRAL = "neutral"
+    """中性"""
+    FEAR = "fear"
+    """恐惧"""
+    SADNESS = "sadness"
+    """悲伤"""
+    ANGER = "anger"
+    """愤怒"""
+    DISGUST = "disgust"
+    """厌恶"""
+    PANIC = "panic"
+    """恐慌"""
+
+    @property
+    def tendency(self) -> EmotionTendency:
+        """获取情绪倾向: POSITIVE(积极), NEGATIVE(消极), NEUTRAL(中性)"""
+        positive = {Emotion.JOY, Emotion.CONTENTMENT}
+        negative = {
+            Emotion.FEAR,
+            Emotion.SADNESS,
+            Emotion.ANGER,
+            Emotion.DISGUST,
+            Emotion.PANIC,
+        }
+
+        if self in positive:
+            return EmotionTendency.POSITIVE
+        elif self in negative:
+            return EmotionTendency.NEGATIVE
+        return EmotionTendency.NEUTRAL
+
+    @property
+    def prompt(self) -> str:
+        """返回情绪提示"""
+        return f"Your current emotion is {self.value}. The output should reflect this emotion. If there is an emotional shift, ensure the transition is natural to simulate human emotional changes."
+
+
+class EmotionAnalysis:
+    """情绪分析类"""
+
+    def __init__(self, provider: ProviderOpenAI):
+        self.provider = provider
+
+    async def analyze(self, text: dict) -> dict | None:
+        """分析文本情绪"""
+        sys_prompt = EMOTION_ANALYSIS_PROMPT.format(Emotion=Emotion)
+        res = await self.provider.text_chat(
+            system_prompt=sys_prompt,
+            prompt=json.dumps(text),
+        )
+
+        try:
+            json_pattern = r"```(?:json)?\s*([\s\S]*?)\s*```"
+            matches = re.findall(json_pattern, res.completion_text)
+
+            if matches:
+                json_content = matches[0]
+            else:
+                json_content = (
+                    res.completion_text.replace("```json", "")
+                    .replace("```", "")
+                    .strip()
+                )
+
+            cleaned_content = json_content.replace("{{", "{").replace("}}", "}").strip()
+            resp = json.loads(cleaned_content)
+
+            return resp
+
+        except json.JSONDecodeError as e:
+            logger.error(f"JSON parsing error: {e!s}")
+            logger.error(f"Failed content: {cleaned_content}")
+            return None
+        except Exception as e:
+            logger.error(f"Unexpected error in emotion analysis: {e!s}")
+            return None

core/util/prompts.py

@@ -151,3 +151,60 @@
 ```
 
 - You should use `USER_ID` as the source or target content for any self-references (e.g., "I", "me", "my" etc.) in user messages."""
+
+EMOTION_ANALYSIS_PROMPT = """
+You are an advanced sentiment analysis AI, specializing in persona-driven emotion detection.
+
+You will be given:
+1. A JSON object with:
+   - "text": "the immediate text to analyze (latest message)"
+   - "personality": "string describing the persona's personality traits"
+   - "context": "previous conversation history (for reference only, to understand the nuance of the current text)"
+
+## Your Task
+1. **Analyze the "text" to determine the speaker's true emotional state**, focusing primarily on the explicit content.
+2. **Deeply consider the persona's traits**: How would this persona typically express or mask emotions? How do their personality, habits, or background influence their emotional expression?
+3. Use **context** (recent conversation, situation) only to clarify ambiguous or implicit emotions, or to resolve sarcasm/irony.
+4. Select from these emotions: {[e.value for e in Emotion]}
+5. Determine intensity ranging from [-1, 1]
+
+## Output Format
+```json
+{{
+  "emotion": "<emotion_name>",
+  "intensity": <float>
+}}
+```
+
+## Analysis Priority
+1. **Text Analysis (Primary Focus):**
+   - What emotion does the speaker actually express in this text?
+   - Emotional vocabulary, tone, punctuation, emojis, and markers.
+   - Intensity and clarity of emotional expression.
+
+2. **Persona Consideration (Secondary, but Critical):**
+   - How does the persona's character shape their emotional display?
+   - Would this persona exaggerate, suppress, or distort certain emotions?
+   - Adjust your judgment based on persona's typical emotional baseline and expression style.
+
+3. **Context Reference (Tertiary):**
+   - Use only to clarify ambiguous, sarcastic, or context-dependent emotions.
+   - Reference recent conversation or situation if it changes the meaning of the text.
+
+## Guidelines
+1. **Emotion Selection:**
+   - Choose from: {[e.value for e in Emotion]}
+   - Focus on the dominant, most likely emotion the speaker is experiencing.
+   - Only use "neutral" if the text is truly emotionless or ambiguous.
+
+2. **Intensity Calculation:**
+   - Consider word choice, expression strength, and emotional markers.
+   - Adjust for persona's typical emotional range and expression habits.
+   - Use context only if it clearly amplifies or dampens the emotion.
+
+3. **Principles:**
+   - Prioritize evidence from the text itself.
+   - Let persona traits guide your interpretation of ambiguous or subtle emotions.
+   - Use context to resolve uncertainty, not as the main basis.
+   - Be objective and avoid over-interpretation.
+"""