Recommender system design and multi-channel pricing: personalization strategies for online platforms

• Main Author: ZHANG, Hao
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2024
• Subjects: Recommender Systems; User-generated Content; Content Generation and Usage; Multi-channel Pricing; Consumer Search; Reference Price; Computer and Systems Architecture
• Online Access: https://ink.library.smu.edu.sg/etd_coll/600; https://ink.library.smu.edu.sg/context/etd_coll/article/1598/viewcontent/GPIS_AY2021_PhD_ZHANG_Hao.pdf
• Institution: Singapore Management University

The advancement of mobile technology and rising consumer demands have contributed to the unprecedented growth of online platforms. In online platforms, recommender systems connect with multistakeholders who have different interests. Designing recommender systems to balance the benefit of multistakeholders is important for these platforms. In addition, price is an important factor influencing consumers’ purchase decisions. An increasing number of online platforms introduce multiple sales channels. Optimizing multiple-channel prices is vital for these platforms. Thus, this thesis designs multistakeholder recommender systems and multi-channel pricing strategies for online platforms through the following two works. The first work focuses on designing multistakeholder recommender systems for balancing creators’ content generation and users’ content usage on two-sided entertainment platforms. Entertainment platforms such as YouTube and Spotify host a vast amount of user-generated content (UGC). The unique feature of UGC entertainment platforms is that creators’ content generation and users’ content usage can influence each other. However, traditional recommender systems often emphasize content usage but ignore content generation, leading to a misalignment between these two goals. To address this challenge, we propose a new prescriptive uplift framework to balance content generation and usage through personalized content recommendation and display. The proposed framework includes three interconnected parts, which are modeling user behaviors, estimating heterogeneous treatment effects (HTEs) on users and creators, and recommending and displaying contents, respectively. Using a large-scale real-world dataset of an entertainment platform, we demonstrate that the proposed recommendation and display method not only balances content generation and usage, but also improves participants’ activity (i.e., the overall level of content generation and usage) by 17%-28% compared to benchmark models. Our work is among the first to balance content generation and usage through recommendation and display decisions, which contributes to the current literature in the areas of content management, recommender systems, and product display. Our method also identifies different types of content generation and usage behaviors and provides important managerial implications for platforms in making personalized content recommendation and display decisions. The second work focuses on designing multi-channel pricing strategies based on consumers’ multi-channel footprints on e-commerce platforms. Consumers’ multi-channel footprints contain rich information about their preferences and search costs, which are very valuable for understanding consumer behaviors and informing platforms’ multi-channel pricing strategies. However, prior studies focus mainly on consumers’ single-channel footprints without considering consumers’ cross-channel search process. To address this challenge, we explore consumers’ multi-channel footprints on e-commerce platforms to optimize platforms’ multi-channel pricing decisions. We first develop a structural model to characterize consumers’ multi-channel footprints as multi-channel sequential search (MSS) behaviors that consist of cross-channel, cross-product, and cross-page searches, where consumers adaptively update their cross-channel reference prices along the search paths. The formed cross-channel reference price effect alters the relative appeal of unsearched products and channels compared to those already searched, which influences consumers’ future search paths and impacts their purchase decisions. Next, using a large-scale dataset from an e-commerce platform, we demonstrate that the proposed MSS model can improve click prediction accuracy by 7% and purchase prediction accuracy by 21% compared to the state-of-the-art sequential search models. Finally, by harnessing consumers’ heterogeneous MSS behaviors and the cross-channel reference price effect, we assist the platform in optimizing personalized multi-channel price promotion decisions, achieving an 18% profit improvement over the existing promotion approach. Our proposed MSS model extends the existing search framework to a multi-channel setting by capturing the interplay between consumers’ cross-channel search and cross-channel reference price effect, which comprehensively characterizes consumers’ multi-channel search behaviors. Our findings provide important managerial and operational implications for e-commerce platforms’ multi-channel pricing strategies.