Method to detect and score churn in online social games

摘要

A method and a system for predicting churn of a player of an online game is described. Online engagements of a group of players of the online game are monitored during a churn prediction model training period. Online engagement scores for the group of players are computed within a periodic number of days within the churn model training period. A weighted exponential moving average of the online engagement scores of the group of players of the online game is computed during the churn model training period. The weighted exponential moving average is used to determine an online engagement threshold value of a churn prediction model for the online game. The online engagement threshold value is applied to a weighted exponential moving average of a player during an observation period to determine a churn probability of the player within a prediction period.

主权项

1. A computer-implemented method comprising:
monitoring online engagements of a group of players of an online game during a training period; using at least one processor to compute respective online engagement scores for the group of players based on how many days each respective player has played the online game within a subset of the training period; assigning respective exponential weights to the online engagement scores so that the respective exponential weights of the online engagement scores increase with an increase in recency within the training period; using the exponential weights to compute a group moving average, the group moving average comprising a weighted exponential moving group average of the online engagement scores of the group of players of the online game during the training period; using the group moving average to determine an online engagement threshold value of a churn prediction model for the online game; and applying the online engagement threshold value to a player moving average to determine a churn probability of a player for a prediction period outside the training period, the player moving average comprising a weighted exponential moving player average of an online engagement score of the player during an observation period within the training period.