Prompt Learning Unlocked for App Promotion in the Wild

Zhongyu Ouyang, Shifu Hou, Shang Ma, Chaoran Chen, Chunhui Zhang, Toby Li, Xusheng Xiao, Chuxu Zhang, Yanfang Ye

Keywords: Prompt Learning; App Promotion Network; Graph Completion

TL;DR: This paper introduces Prompt Promotion, a transformer-based framework, in the context of app promotion, that unlocks prompt learning capabilities by incorporating metapath- and embedding-based prompts to provide hints for pattern prediction guidance.

Abstract: In recent times, mobile apps have increasingly incorporated app promotion ads to promote other apps, raising cybersecurity and online commerce concerns related to societal trust and recommendation systems. To effectively discover the intricate nature of the app promotion graph data, we center around the graph completion task, aiming to learn the connection patterns among diverse relations and entities. However, accurately deciphering the connection patterns in such a large and diverse graph presents significant challenges for deep learning models. To overcome these challenges, we introduce Prompt Promotion, a transformer-based framework that unlocks prompt learning capabilities by incorporating metapath- and embedding-based prompts that provide valuable hints to guide the model's predictions for undetermined connection patterns. Experimental results show that our Prompt Promotion model represents a pioneering prompt-based capability in effectively completing the app promotion graph. It not only demonstrates superior performance in heterogeneous graph completion in real-world scenarios, but also exhibits strong generalization capabilities for diverse, complex, and noisy connection patterns when paired with their respective prompts.

Submission Number: 9