Smart Approval of Gaming API Applications Through BERT and MLP Classification Models

This work addresses the challenge of enhancing the user experience within Riot Games' developer portal by streamlining the process of creating secure and efficient applications that interact with its suite of Application Programming Interfaces (APIs). To accomplish this, a spam detection model was developed using a hybrid deep learning approach that combines Bidirectional Encoder Representations from Transformers (BERT) for semantic understanding and a Multilayer Perceptron (MLP) for classification. T he p roposed m odel a nalyzes t extual metadata submitted with each application to predict whether it should be approved or rejected. Experimental results demonstrate strong classification performance, with an a ccuracy of 91.34%, an F1score of 91.69%, and a Receiver Operating Characteristic - Area Under the Curve (ROC AUC) of 91.33%. These metrics indicate the model's effectiveness in minimizing both false positives and false negatives. Furthermore, the consistent decrease observed in both training and validation loss across epochs reflects stable convergence and robust generalization capabilities. Specifically, this work lies in the deployment of a lightweight, domainspecific s pam d etection p ipeline t ailored t o t he u nique context of game development APIs. This system not only accelerates the application approval process but also enhances platform integrity by proactively identifying and filtering p otentially m alicious or low-quality submissions. The proposed solution paves the way for more secure, efficient, and developer-friendly gaming ecosystems.