Abstract:With the rapid prevalence and explosive development of MOBA esports (Multiplayer Online Battle Arena electronic sports), much research effort has been devoted to automatically predicting game results (win predictions). While this task has great potential in various applications, such as esports live streaming and game commentator AI systems, previous studies fail to investigate the methods to interpret these win predictions. To mitigate this issue, we collected a large-scale dataset that contains real-time game records with rich input features of the popular MOBA game Honor of Kings. For interpretable predictions, we proposed a Two-Stage Spatial-Temporal Network (TSSTN) that can not only provide accurate real-time win predictions but also attribute the ultimate prediction results to the contributions of different features for interpretability. Experiment results and applications in real-world live streaming scenarios showed that the proposed TSSTN model is effective both in prediction accuracy and interpretability.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the interpretability issue of real-time win rate prediction in MOBA esports (such as "Honor of Kings"). #### Background and Motivation - **Background**: With the rapid popularity and development of MOBA esports (Multiplayer Online Battle Arena esports), many studies have focused on automatically predicting game outcomes (win rate prediction). These predictions have great potential applications in esports live streaming and commentary systems. - **Motivation**: Although existing research has made progress in prediction accuracy, there is a lack of interpretability of the prediction results. For example, during the live broadcast of the 2019 League of Legends World Championship, non-explanatory prediction models could only provide the win rates of the two teams (e.g., 80%) without explaining the reasons behind them, leading commentators to make only simple comments. #### Main Issues - **Problem Description**: Existing research on win rate prediction in MOBA esports mainly focuses on prediction accuracy but neglects how to interpret these prediction results. This limits its application in real-world scenarios. - **Specific Problem**: How to design a model that can provide both accurate real-time win rate predictions and explain the reasons behind the prediction results? ### Solution - **Dataset Collection**: Researchers collected a large-scale "Honor of Kings" dataset, which includes rich real-time game records and feature information. These data come from matches of high-level players, making them more reliable. - **Model Design**: A Two-Stage Spatial-Temporal Network (TSSTN) is proposed, which not only provides accurate real-time win rate predictions but also explains the feature contributions behind the prediction results. - **First Stage (Spatial Stage)**: Features are divided into six different feature groups (gold, kills, turrets, jungle resources, minions, heroes), and an independent spatial model is built for each feature group to make predictions. - **Second Stage (Temporal Stage)**: The outputs of different spatial models are combined through time-varying weights to obtain the final prediction result and explain the importance and contribution of each feature group. ### Experimental Validation - **Experimental Setup**: Experiments were conducted using the collected dataset, with 10,000 matches randomly selected as the test set, and the remaining part used for training and validation. - **Comparison Methods**: - **Heuristic**: Prediction based on the gold difference accumulated by the two teams. - **Fully-Connected**: A fully connected neural network that inputs all feature groups for prediction. - **Results**: Experimental results show that the TSSTN model outperforms other baseline models in both prediction accuracy and interpretability. ### Conclusion - This paper addresses the interpretability issue of real-time win rate prediction in MOBA esports by designing the TSSTN model, which not only improves prediction accuracy but also provides detailed explanatory information, aiding the application in esports live streaming and commentary systems.

Interpretable Real-Time Win Prediction for Honor of Kings, a Popular Mobile MOBA Esport

Predicting Events in MOBA Games: Prediction, Attribution, and Evaluation

DotaNet: Two-Stream Match-Recurrent Neural Networks for Predicting Social Game Result

Match Tracing: A Unified Framework for Real-time Win Prediction and Quantifiable Performance Evaluation

Predicting Outcomes in Video Games with Long Short Term Memory Networks

The Application of Machine Learning in Predicting the Results of Popular eSports Games: Win Rate Prediction in MOBA and FPS Games

Real-Time Strategy Game Win-Lose Prediction Model Using 3D Convolutional Neural Networks

Machine learning models for DOTA 2 outcomes prediction

An Uncertainty-Incorporated Approach to Predict the Winner in StarCraft II Using Neural Processes

Outcome Prediction Of Dota2 Using Machine Learning Methods

Win Prediction in Esports: Mixed-Rank Match Prediction in Multi-player Online Battle Arena Games

A Modified Multi-size Convolution Neural Network for Winner Prediction Based on Time Serial Datasets

Predicting Game Outcome in Multiplayer Online Battle Arena Games

Multi-Dimensional Prediction of Guild Health in Online Games: A Stability-Aware Multi-Task Learning Approach

Real-Time Crash Risk Prediction using Long Short-Term Memory Recurrent Neural Network

Categorical Learning-based Line-up Prediction in the Drafting Process of MOBA Games

A Multi-Size Convolution Neural Network for RTS Games Winner Prediction

Hierarchical Attention Based Recurrent Neural Network Framework for Mobile MOBA Game Recommender Systems

A Machine Learning Approach to Detect Strategic Behavior from Large-Population Observational Data Applied to Game Mode Prediction on a Team-Based Video Game

MOBA game trend prediction model based on sequence-to-sequence structure

Realtime mobile bandwidth prediction using LSTM neural network and Bayesian fusion