Paper Type
full
Description
Digitally enabled two-sided platforms rely on mediating different actors to evoke transactions. Here, the core value-generating mechanisms of these platforms relate to the recommendations that persuade users to make future transactions, thereby driving sales, customer satisfaction, efficiency, and trust. To generate effective recommendations, accurate user profiling is fundamental. Ubiquitous computing provides valuable data to enhance user profiling by uncovering behavioral patterns of individual users. Specifically, through machine learning methods, recommendation systems are able to understand users better by considering both past individual behaviors and their respective contexts. However, state-of-the-art recommendation systems rely either on collaborative or content-based approaches, thereby neglecting a user’s time-varying contexts and the dynamics that influence these contexts. We address this shortcoming by developing a context-aware and user-specific hybrid recommendation system using transfer-learning techniques based on (recurrent) neural networks. Evaluating our approach on Expedia’s hotel booking data, we demonstrate its enhanced performance compared to common recommendation approaches.
Recommended Citation
Tofangchi, Schahin; Hanelt, André; and Li, Siyuan, "Advancing Recommendations on Two-Sided Platforms: A Machine Learning Approach to Context-Aware Profiling" (2019). ICIS 2019 Proceedings. 21.
https://aisel.aisnet.org/icis2019/data_science/data_science/21
Advancing Recommendations on Two-Sided Platforms: A Machine Learning Approach to Context-Aware Profiling
Digitally enabled two-sided platforms rely on mediating different actors to evoke transactions. Here, the core value-generating mechanisms of these platforms relate to the recommendations that persuade users to make future transactions, thereby driving sales, customer satisfaction, efficiency, and trust. To generate effective recommendations, accurate user profiling is fundamental. Ubiquitous computing provides valuable data to enhance user profiling by uncovering behavioral patterns of individual users. Specifically, through machine learning methods, recommendation systems are able to understand users better by considering both past individual behaviors and their respective contexts. However, state-of-the-art recommendation systems rely either on collaborative or content-based approaches, thereby neglecting a user’s time-varying contexts and the dynamics that influence these contexts. We address this shortcoming by developing a context-aware and user-specific hybrid recommendation system using transfer-learning techniques based on (recurrent) neural networks. Evaluating our approach on Expedia’s hotel booking data, we demonstrate its enhanced performance compared to common recommendation approaches.