AI-система аналитики поведения игроков
Аналитика игроков — ключевой инструмент монетизации и удержания в gaming. Когда игрок перестаёт прогрессировать, злится или скучает — это поведенческие сигналы, видимые в данных за минуты до churn. AI-система детектирует их в реальном времени и запускает персонализированные интервенции.
Профилирование игрока и сегментация
import numpy as np
import pandas as pd
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import GradientBoostingClassifier
class PlayerProfiler:
"""Многомерное профилирование игрока"""
def extract_behavioral_features(self, sessions: pd.DataFrame,
player_id: str,
window_days: int = 30) -> dict:
"""Признаки из игровых сессий за период"""
player_sessions = sessions[
(sessions['player_id'] == player_id) &
(sessions['date'] >= pd.Timestamp.now() - pd.Timedelta(days=window_days))
]
if player_sessions.empty:
return {}
return {
# Активность
'sessions_per_week': len(player_sessions) / (window_days / 7),
'avg_session_minutes': player_sessions['duration_minutes'].mean(),
'total_play_hours': player_sessions['duration_minutes'].sum() / 60,
'days_active': player_sessions['date'].dt.date.nunique(),
# Прогресс
'avg_level_gain_per_session': player_sessions['level_gained'].mean(),
'completion_rate': player_sessions['objective_completed'].mean(),
'win_rate': player_sessions['wins'].sum() / max(player_sessions['games_played'].sum(), 1),
# Монетизация
'total_spent_usd': player_sessions['purchase_usd'].sum(),
'purchase_count': (player_sessions['purchase_usd'] > 0).sum(),
'avg_purchase_usd': player_sessions[player_sessions['purchase_usd'] > 0]['purchase_usd'].mean() or 0,
# Социальность
'social_interactions': player_sessions['chat_messages'].sum() + player_sessions['party_plays'].sum(),
'guild_member': int(player_sessions['guild_id'].notna().any()),
# Разнообразие
'game_modes_played': player_sessions['game_mode'].nunique(),
'avg_frustration_events': player_sessions.get('rage_quits', pd.Series([0])).mean(),
}
def segment_players(self, player_features: pd.DataFrame,
n_segments: int = 6) -> pd.DataFrame:
"""K-Means сегментация игроков"""
feature_cols = [c for c in player_features.columns if c != 'player_id']
X = player_features[feature_cols].fillna(0)
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
kmeans = KMeans(n_clusters=n_segments, random_state=42, n_init=10)
player_features['segment'] = kmeans.fit_predict(X_scaled)
# Интерпретация сегментов по центроидам
centroids = pd.DataFrame(
scaler.inverse_transform(kmeans.cluster_centers_),
columns=feature_cols
)
segment_labels = self._label_segments(centroids)
player_features['segment_label'] = player_features['segment'].map(segment_labels)
return player_features
def _label_segments(self, centroids: pd.DataFrame) -> dict:
"""Автоматическая маркировка сегментов по характеристикам"""
labels = {}
for i, row in centroids.iterrows():
spend = row.get('total_spent_usd', 0)
sessions = row.get('sessions_per_week', 0)
social = row.get('social_interactions', 0)
if spend > 50 and sessions > 10:
labels[i] = 'whale'
elif spend > 20:
labels[i] = 'dolphin'
elif sessions > 14:
labels[i] = 'hardcore_f2p'
elif social > 100:
labels[i] = 'social_player'
elif sessions < 2:
labels[i] = 'casual_at_risk'
else:
labels[i] = 'regular'
return labels
class ChurnPredictor:
"""Предсказание оттока игроков"""
def __init__(self):
self.model = GradientBoostingClassifier(
n_estimators=200, learning_rate=0.05, max_depth=4, random_state=42
)
def build_churn_features(self, player_history: pd.DataFrame) -> pd.DataFrame:
"""Признаки, предсказывающие churn в горизонте 14 дней"""
features = pd.DataFrame()
features['sessions_last_7d'] = player_history['sessions_7d']
features['sessions_prev_7d'] = player_history['sessions_prev_7d']
features['session_trend'] = (
(features['sessions_last_7d'] - features['sessions_prev_7d']) /
(features['sessions_prev_7d'] + 1)
)
features['days_since_last_login'] = player_history['days_since_last_login']
features['avg_session_drop_min'] = player_history['avg_session_drop_minutes']
features['level_progression_rate'] = player_history['level_gain_per_hour']
features['win_rate_last_10'] = player_history['win_rate_last_10_games']
features['frustration_events'] = player_history['rage_quits_7d']
features['purchase_recency_days'] = player_history['days_since_last_purchase']
features['social_activity_trend'] = player_history['social_activity_trend']
return features.fillna(0)
def predict_churn_risk(self, players: pd.DataFrame) -> pd.DataFrame:
"""Скор риска оттока для каждого игрока"""
X = self.build_churn_features(players)
churn_probabilities = self.model.predict_proba(X)[:, 1]
result = players[['player_id']].copy()
result['churn_probability_14d'] = churn_probabilities
result['churn_risk'] = pd.cut(
churn_probabilities,
bins=[0, 0.2, 0.5, 0.75, 1.0],
labels=['low', 'medium', 'high', 'critical']
)
return result
class PlayerInterventionEngine:
"""Персонализированные интервенции для удержания"""
INTERVENTIONS = {
'whale': {
'critical': {'type': 'vip_outreach', 'channel': 'personal_manager'},
'high': {'type': 'exclusive_content', 'channel': 'in_game_popup'},
},
'regular': {
'critical': {'type': 'comeback_bonus', 'channel': 'push_notification'},
'high': {'type': 'limited_offer', 'channel': 'email'},
},
'hardcore_f2p': {
'critical': {'type': 'challenge_event', 'channel': 'in_game'},
'high': {'type': 'new_content_unlock', 'channel': 'push_notification'},
},
'casual_at_risk': {
'critical': {'type': 'simplified_quest', 'channel': 'push_notification'},
'high': {'type': 'social_invite', 'channel': 'email'},
}
}
def select_intervention(self, player_segment: str,
churn_risk: str,
player_context: dict) -> dict:
segment_interventions = self.INTERVENTIONS.get(player_segment, self.INTERVENTIONS['regular'])
intervention = segment_interventions.get(churn_risk, {'type': 'generic_reminder', 'channel': 'push'})
# Персонализация контента интервенции
intervention['personalized_offer'] = self._create_offer(
player_segment, player_context
)
return intervention
def _create_offer(self, segment: str, context: dict) -> dict:
offers = {
'whale': {'type': 'exclusive_skin', 'value': 'Limited Edition Character'},
'dolphin': {'type': 'currency_bonus', 'value': '50% bonus gems'},
'hardcore_f2p': {'type': 'xp_boost', 'value': '2x XP for 3 days'},
'casual_at_risk': {'type': 'starter_pack', 'value': 'Welcome back pack'},
}
return offers.get(segment, {'type': 'general_bonus', 'value': 'Daily reward'})
Предсказание churn в gaming достигает AUC 0.82-0.88 с 7-дневным горизонтом. Персонализированные интервенции для сегмента «высокий риск» удерживают 25-35% потенциально уходящих игроков. Whale-сегмент (5-10% игроков) генерирует 50-70% revenue — приоритет при скоринге критических.