A Semi-Supervised and Inductive Embedding Model for Churn Prediction of Large-Scale Mobile Games

Abstract

Mobile gaming has emerged as a promising market with billion-dollar revenues. A variety of mobile game platforms and services have been developed around the world. One critical challenge for these platforms and services is to understand user churn behavior in mobile games. Accurate churn prediction will benefit many stakeholders such as game developers, advertisers, and platform operators. In this paper, we present the first largescale churn prediction solution for mobile games. In view of the common limitations of the state-of-the-art methods built upon traditional machine learning models, we devise a novel semisupervised and inductive embedding model that jointly learns the prediction function and the embedding function for userapp relationships. We model these two functions by deep neural networks with a unique edge embedding technique that is able to capture both contextual information and relationship dynamics. We also design a novel attributed random walk technique that takes into consideration both topological adjacency and attribute similarities. To evaluate the performance of our solution, we collect real-world data from the Samsung Game Launcher platform that includes tens of thousands of games and hundreds of millions of user-app interactions. The experimental results with this data demonstrate the superiority of our proposed model against existing state-of-the-art methods.

Real-Time Streaming Graph Embedding Through Local Actions

“…This property is referred to as temporal smoothness. This property has also been observed and shown to be helpful to improve representation performance in [5]. Suppose that we are at time t + 1.…”

“…There are several studies on learning representations in dynamic graphs, but very few in streaming graphs, which require higher efficiency and lower uncertainty. Liu et al [5] and Hamilton et al [15] propose to leverage node feature information to learn an embedding function that generalizes to unseen nodes. These two methods rely on prior knowledge of new vertices' attributes and are difficult to work with only topology information.…”

“…The learned graph representations are regarded as informative feature inputs to various machine learning tasks. Graph representation learning has been proven to be a useful tool for many machine learning tasks, such as vertex classification [2], community detection [3], link reconstruction [4], and dis-link prediction [5].…”

Micro- and macro-level churn analysis of large-scale mobile games

“…We tune the hyperparameters, including learning rate, batch size, regularization terms, number of layers and number of neurons per layer, based on the model performance on the test datasets. To determine the parameters for micro-level churn prediction, we follow what we did in our previous work [28]. A grid search on these parameters is performed and the combination yielding the best performance is chosen.…”

Modelling Early User-Game Interactions for Joint Estimation of Survival Time and Churn Probability

“…On top of what is illustrated so far, we also individuated a series of limitations regarding the employed data-sources. The number of the considered users rarely goes beyond 10 4 [24] and when it comes to churn estimation the class distribution is usually greatly imbalanced, both of these factors can pose limitations on the interpretation and generalization of results.…”