feat(social): add listening activity, shared rooms and modern chat

services/recommender/pipu_rec/wave/engine.py

@@ -313,6 +313,7 @@ def build_batch(
     rng: np.random.Generator | None = None,
     ranker: Ranker | None = None,
     ctx: Context | None = None,
+    friend_ids: list[str] | None = None,
 ) -> list[Pick]:
     """Кандидаты → фильтры → оценка → соответствие контексту → MMR с правилами разнообразия."""
     rng = rng or np.random.default_rng()
@@ -363,6 +364,8 @@ def build_batch(
         if seeds:
             q = index.emb[seeds].mean(axis=0)
             add(_top((index.emb @ q).astype(np.float64), allowed & index.has_emb, 100), "seed")
+    friends = [index.pos[t] for t in (friend_ids or []) if t in index.pos and allowed[index.pos[t]]][:10]
+    add(friends, "friends")
     query = np.maximum(sim_session, sim_taste) if svec is not None or len(centroids) else None
     if query is not None:
         # «Новое» помечаем раньше audio_sim: неслышанный трек рядом со вкусом считается исследованием.
@@ -446,6 +449,8 @@ def build_batch(
         pred = ranker.predict(x)
         score = np.argsort(np.argsort(pred)) / max(len(pred) - 1, 1)
 
+    # Social influence is small, context-sensitive, and survives the learned ranker.
+    score = score + 0.06 * np.isin(cand, friends) * f
     # 5. MMR + правила: 1 трек исполнителя на пачку, не больше 2 на последние 10; доля «нового».
     picks: list[Pick] = []
     chosen: list[int] = []
@@ -457,6 +462,8 @@ def build_batch(
         best_j, best_v = -1, -np.inf
         for j in order:
             i = int(cand[j])
+            if sources[i] == "friends" and any(p.source == "friends" for p in picks):
+                continue
             a = index.artist[i]
             if any(index.artist[c] == a for c in chosen) or recent_artists.count(a) >= 2:
                 continue
@@ -471,7 +478,9 @@ def build_batch(
                 continue
             if chosen:
                 chosen_artists = {index.artist[c] for c in chosen}
-                rest = [j for j in order if index.artist[int(cand[j])] in chosen_artists]
+                social_chosen = any(p.source == "friends" for p in picks)
+                rest = [j for j in order if index.artist[int(cand[j])] in chosen_artists
+                        and not (sources[int(cand[j])] == "friends" and social_chosen)]
                 if not rest:
                     break
                 best_j = max(rest, key=lambda j: score[j])

services/recommender/pipu_rec/wave/service.py

@@ -38,6 +38,7 @@ class BatchRequest(BaseModel):
     mood: str | None = None
     seed: Seed | None = None
     exclude_ids: list[str] = Field(default_factory=list, max_length=1000)
+    friend_track_ids: list[str] = Field(default_factory=list, max_length=30)
     feedback: list[Feedback] = Field(default_factory=list, max_length=200)
     n: int = Field(default=5, ge=1, le=20)
 
@@ -93,6 +94,7 @@ class WaveService:
             n=req.n,
             exclude=set(req.exclude_ids),
             seed_ids=seeds,
+            friend_ids=req.friend_track_ids,
             rng=np.random.default_rng(),
             ranker=ranker,
             ctx=Context(has_activity=bool(req.activity), has_mood=bool(req.mood)),

services/recommender/tests/test_engine.py

@@ -45,6 +45,23 @@ def profile(**kw: object) -> Profile:
     return Profile(**base)  # type: ignore[arg-type]
 
 
+def test_friend_signal_is_bounded_and_respects_negative_feedback() -> None:
+    index = library()
+    blocked = {"t0-0", "t1-0"}
+    picks = build_batch(index, profile(disliked=blocked), Session(), [], n=5,
+                        friend_ids=index.ids, rng=np.random.default_rng(8))
+    assert not blocked.intersection(p.id for p in picks)
+    assert sum(p.source == "friends" for p in picks) <= 1
+    assert len(picks) == 5
+
+
+def test_empty_friend_signal_does_not_change_ranking() -> None:
+    index = library()
+    first = build_batch(index, profile(), Session(), [], rng=np.random.default_rng(17))
+    second = build_batch(index, profile(), Session(), [], friend_ids=[], rng=np.random.default_rng(17))
+    assert [p.id for p in first] == [p.id for p in second]
+
+
 def cluster(track_id: str) -> int:
     return int(track_id[1])