Paper Number
ICIS2025-1054
Paper Type
Complete
Abstract
Live streaming is a growing industry that faces unique recommendation challenges due to the disparity in user behaviors, such as high viewing rates versus low gifting rates. To address this, we propose the Parallel Insight Network (PIN), a novel model that leverages both abundant viewing data and sparse gifting data to improve gifting-oriented recommendations. Our key contribution is an architecture designed to mitigate data sparsity issues and enhance recommendations through parallel self-attention mechanisms and knowledge transfer techniques. The model achieves superior results across multiple recommendation tasks including overall accuracy, channel discovery, cold-starts, and freemium-to-premium conversion prediction. Evaluated on real-world data with a sample size of 156,700 users, it demonstrates a 6.1%–131.1% improvement in gifting prediction accuracy. These results highlight the benefits of leveraging heterogeneous data on freemium-based platforms, contributing to both freemium business model literature and recommender systems research.
Recommended Citation
Lu, Xintong; Bi, Xuan; Guo, Yue; and Chong, Alain, "Turning Freemium into Gold: A Parallel Insight Network for Live Streaming Recommendation" (2025). ICIS 2025 Proceedings. 1.
https://aisel.aisnet.org/icis2025/da_bus/da_bus/1
Turning Freemium into Gold: A Parallel Insight Network for Live Streaming Recommendation
Live streaming is a growing industry that faces unique recommendation challenges due to the disparity in user behaviors, such as high viewing rates versus low gifting rates. To address this, we propose the Parallel Insight Network (PIN), a novel model that leverages both abundant viewing data and sparse gifting data to improve gifting-oriented recommendations. Our key contribution is an architecture designed to mitigate data sparsity issues and enhance recommendations through parallel self-attention mechanisms and knowledge transfer techniques. The model achieves superior results across multiple recommendation tasks including overall accuracy, channel discovery, cold-starts, and freemium-to-premium conversion prediction. Evaluated on real-world data with a sample size of 156,700 users, it demonstrates a 6.1%–131.1% improvement in gifting prediction accuracy. These results highlight the benefits of leveraging heterogeneous data on freemium-based platforms, contributing to both freemium business model literature and recommender systems research.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
07-DataAnalytics