JARVIS/backend/app/services/memory/emotion_analyzer.py

"""
EmotionAnalyzer

Extracts emotional intensity from text and calculates emotion-based importance scores.
"""

import re
from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from app.models.memory import UserMemory


class EmotionAnalyzer:
    """Analyze emotional keywords in memory content"""

    # Emotion keyword weights (higher = more important)
    EMOTION_KEYWORDS = {
        # High intensity
        "急": 1.0,
        "紧急": 1.0,
        "很重要": 0.9,
        "非常重要": 1.0,
        "必须": 0.9,
        # Medium-high intensity
        "困扰": 0.8,
        "烦恼": 0.7,
        "担心": 0.7,
        "焦虑": 0.8,
        "害怕": 0.8,
        "恐惧": 0.9,
        # Medium intensity
        "想解决": 0.6,
        "希望": 0.5,
        "想要": 0.4,
        "需要": 0.4,
        "渴望": 0.6,
        # Low intensity (casual/neutral)
        "无所谓": 0.1,
        "随便": 0.1,
        "都行": 0.1,
        "还好": 0.2,
        # Negative valence
        "讨厌": 0.6,
        "不喜欢": 0.5,
        "恨": 0.8,
        "不喜欢": 0.5,
        # Positive valence
        "喜欢": 0.5,
        "爱": 0.7,
        "开心": 0.4,
        "高兴": 0.4,
    }

    # Urgency patterns — only match when there's an explicit time-bound word
    # NOTE: "今天" alone is too common and matches neutral sentences like "今天天气不错"
    URGENCY_PATTERNS = [
        (re.compile(r"马上|立刻|立即|赶紧"), 1.0),
        (re.compile(r"今天内|今天必须|今日必须"), 0.8),
        (re.compile(r"明天|明天之前|明日"), 0.6),
        (re.compile(r"这周|本周"), 0.4),
        (re.compile(r"尽快|早点"), 0.7),
    ]

    def extract(self, text: str) -> list[str]:
        """Extract emotion keywords from text

        Returns:
            List of matched emotion keywords
        """
        if not text:
            return []

        matched = []
        text_lower = text.lower()

        for keyword, weight in self.EMOTION_KEYWORDS.items():
            if keyword in text_lower:
                matched.append(keyword)

        # Check urgency patterns
        for pattern, weight in self.URGENCY_PATTERNS:
            if pattern.search(text):
                matched.append(f"[URGENCY:{weight}]")

        return matched

    def calculate_score(self, memory: "UserMemory") -> float:
        """Calculate emotion-based importance score (0.0 - 1.0)

        Uses the highest-weighted emotion keyword found in the content.
        """
        content = memory.content or ""
        emotion_tags = memory.emotion_tags or []

        # Check emotion_tags first (pre-extracted)
        if emotion_tags:
            max_weight = 0.0
            for tag in emotion_tags:
                if tag in self.EMOTION_KEYWORDS:
                    max_weight = max(max_weight, self.EMOTION_KEYWORDS[tag])
            if max_weight > 0:
                return max_weight

        # Extract from content
        matched = self.extract(content)
        if not matched:
            return 0.0

        # Get highest weight
        max_weight = 0.0
        for keyword in matched:
            if keyword.startswith("[URGENCY:"):
                # Extract urgency weight
                try:
                    weight = float(keyword.split(":")[1].rstrip("]"))
                    max_weight = max(max_weight, weight)
                except (ValueError, IndexError):
                    pass
            elif keyword in self.EMOTION_KEYWORDS:
                max_weight = max(max_weight, self.EMOTION_KEYWORDS[keyword])

        return min(1.0, max_weight)

    def get_emotion_profile(self, text: str) -> dict:
        """Get detailed emotion profile for text

        Returns:
            Dict with matched keywords, max_weight, and sentiment
        """
        matched = self.extract(text)
        weights = []
        for keyword in matched:
            if keyword.startswith("[URGENCY:"):
                try:
                    weights.append(float(keyword.split(":")[1].rstrip("]")))
                except (ValueError, IndexError):
                    pass
            elif keyword in self.EMOTION_KEYWORDS:
                weights.append(self.EMOTION_KEYWORDS[keyword])

        return {
            "matched_keywords": matched,
            "max_weight": max(weights) if weights else 0.0,
            "avg_weight": sum(weights) / len(weights) if weights else 0.0,
            "sentiment": "positive"
            if (weights and sum(weights) / len(weights) > 0.5)
            else "neutral",
        }