ranking.py•2.32 kB
"""
Ranking profiles for search result scoring.
"""
from dataclasses import dataclass
@dataclass
class RankingProfile:
"""
Ranking parameters for search result scoring.
Attributes:
decay_half_life_hours: Time for recency score to drop by half
decay_weight: Weight for time decay in final score
similarity_weight: Weight for semantic similarity in final score
confidence_weight: Weight for confidence score in final score
"""
decay_half_life_hours: float = 168 # 1 week default
decay_weight: float = 0.33
similarity_weight: float = 0.34
confidence_weight: float = 0.33
class RankingProfiles:
"""Pre-defined ranking profiles for common use cases."""
RECENT_PRIORITY = RankingProfile(
decay_half_life_hours=24, # 1 day half-life
decay_weight=0.6,
similarity_weight=0.3,
confidence_weight=0.1
)
"""Prioritize recent messages (good for debugging, current status)"""
QUALITY_PRIORITY = RankingProfile(
decay_half_life_hours=720, # 30 days half-life
decay_weight=0.1,
similarity_weight=0.4,
confidence_weight=0.5
)
"""Prioritize high-confidence messages (good for proven solutions)"""
BALANCED = RankingProfile(
decay_half_life_hours=168, # 1 week half-life
decay_weight=0.33,
similarity_weight=0.34,
confidence_weight=0.33
)
"""Balanced weighting of all factors (default)"""
SIMILARITY_ONLY = RankingProfile(
decay_half_life_hours=8760, # 1 year (minimal decay)
decay_weight=0.0,
similarity_weight=1.0,
confidence_weight=0.0
)
"""Pure semantic similarity (good for exact topic match)"""
@classmethod
def get_profile(cls, name: str) -> RankingProfile:
"""
Get a named ranking profile.
Args:
name: Profile name ('recent', 'quality', 'balanced', 'similarity')
Returns:
RankingProfile instance
"""
profiles = {
'recent': cls.RECENT_PRIORITY,
'quality': cls.QUALITY_PRIORITY,
'balanced': cls.BALANCED,
'similarity': cls.SIMILARITY_ONLY
}
return profiles.get(name.lower(), cls.BALANCED)