refactor: update profile constants and enhance recommendation logic

Author: TimilsinaBimal

app/services/profile/constants.py

@@ -48,12 +48,14 @@
 FREQUENCY_MULTIPLIER_LOG_FACTOR: Final[float] = 0.1  # Subtle boost
 
 # Top Picks Caps (diversity constraints)
-TOP_PICKS_RECENCY_CAP: Final[float] = 0.10  # Max 10% recent items (from trending/popular)
-TOP_PICKS_GENRE_CAP: Final[float] = 0.40  # Max 40% per genre
+TOP_PICKS_RECENCY_CAP: Final[float] = 0.15  # Max 15% recent items (from trending/popular)
+TOP_PICKS_GENRE_CAP: Final[float] = 0.50  # Max 50% per genre
 TOP_PICKS_CREATOR_CAP: Final[int] = 3  # Max 3 items per creator (director/actor)
-TOP_PICKS_ERA_CAP: Final[float] = 0.40  # Max 40% per era
-TOP_PICKS_MIN_VOTE_COUNT: Final[int] = 500  # Minimum vote count for quality
-TOP_PICKS_MIN_RATING: Final[float] = 7.5  # Minimum weighted rating for quality
+TOP_PICKS_ERA_CAP: Final[float] = 0.50  # Max 50% per era
+TOP_PICKS_MIN_VOTE_COUNT: Final[int] = 250  # Lower noise filter
+TOP_PICKS_MIN_RATING: Final[float] = 7.2  # Minimum weighted rating
+
+MAXIMUM_POPULARITY_SCORE: Final[float] = 15
 
 # Genre whitelist limit (top N genres)
 GENRE_WHITELIST_LIMIT: Final[int] = 7

app/services/profile/sampling.py

@@ -69,7 +69,7 @@ def sample_items(
             if not (it.get("_is_loved") or it.get("_is_liked") or it.get("_id") in added_item_ids)
         ]
 
-        # Always include strong signal items: Loved/Liked: 45%, Added: 20%
+        # Always include strong signal items: Loved/Liked: 30%, Added: 20%
         strong_signal_items = loved_liked_items[: int(max_items * 0.45)] + added_items[: int(max_items * 0.20)]
         strong_signal_scored = [self.scoring_service.process_item(it) for it in strong_signal_items]
 

app/services/recommendation/scoring.py

@@ -4,6 +4,7 @@
 from typing import Any
 
 from app.core.constants import DEFAULT_MINIMUM_RATING_FOR_THEME_BASED_MOVIE, DEFAULT_MINIMUM_RATING_FOR_THEME_BASED_TV
+from app.services.profile.constants import MAXIMUM_POPULARITY_SCORE
 
 
 class RecommendationScoring:
@@ -91,31 +92,21 @@ def m_raw(year: int | None) -> float:
         return m_raw, alpha
 
     @staticmethod
-    def apply_quality_adjustments(score: float, wr: float, vote_count: int, is_ranked: bool, is_fresh: bool) -> float:
-        """Apply multiplicative adjustments based on item quality and source."""
-        q_mult = 1.0
-        if vote_count < 50:
-            q_mult *= 0.75
-        elif vote_count < 150:
-            q_mult *= 0.90
-
-        if wr < 5.5:
-            q_mult *= 0.5
-        elif wr < 6.0:
-            q_mult *= 0.7
-        elif wr >= 7.0 and vote_count >= 500:
-            q_mult *= 1.10
-
-        if is_ranked:
-            if wr >= 6.5 and vote_count >= 200:
-                q_mult *= 1.25
-            elif wr >= 6.0 and vote_count >= 100:
-                q_mult *= 1.10
-
-        if is_fresh and wr >= 7.0 and vote_count >= 300:
-            q_mult *= 1.10
-
-        return score * q_mult
+    def apply_quality_adjustments(
+        score: float, wr: float, vote_count: int, popularity: float, is_ranked: bool, is_fresh: bool
+    ) -> float:
+        """Apply simple quality boost for high-confidence items only."""
+        # Simplified: only boost proven high-quality items, no penalties
+        # Trust the weighted rating formula to handle low vote counts naturally
+        if vote_count >= 1000 and wr >= 7.5 and popularity <= MAXIMUM_POPULARITY_SCORE:
+            # Proven gem: high confidence, high quality
+            return score * 1.10
+        elif vote_count >= 500 and wr >= 7.0 and popularity <= MAXIMUM_POPULARITY_SCORE:
+            # Good confidence and quality
+            return score * 1.05
+
+        # Everything else: trust the base scoring
+        return score
 
     @staticmethod
     def calculate_final_score(
@@ -158,13 +149,14 @@ def calculate_final_score(
         )
         quality_score = RecommendationScoring.normalize(wr)
 
-        # Apply quality adjustments
+        # Simple weighted combination: profile match is primary, quality ensures no bad items
+        base_score = (profile_score * 0.70) + (quality_score * 0.30)
+
+        # light boost for high-confidence items (no penalties!)
         vote_count = item.get("vote_count", 0)
-        adjusted_profile_score = RecommendationScoring.apply_quality_adjustments(
-            profile_score, wr, vote_count, is_ranked=is_ranked, is_fresh=is_fresh
+        popularity = item.get("popularity", 0)
+        final_score = RecommendationScoring.apply_quality_adjustments(
+            base_score, wr, vote_count, popularity, is_ranked=is_ranked, is_fresh=is_fresh
         )
 
-        # Combined score: profile similarity (with quality adjustments) + quality
-        final_score = (adjusted_profile_score * 0.6) + (quality_score * 0.4)
-
         return final_score