Creator-Side Recommender System: Challenges, Designs, and Applications Conference: Companion Proceedings of the ACM Web Conference 2025; April 28-May 2, 2025; Sydney, NSW, Australia Companion Proceedings of the ACM Web Conference 2025 (WWW Companion ’25), April 28-May 2, 2025, Sydney, NSW, Australia

Xiaoshuang Chen, Yibo Wang, Yao Wang, Husheng Liu, Kaiqiao Zhan, Ben Wang, Kun Gai · 2025

Read Creator-Side Recommender System: Challenges, Designs, and Applications Conference: Companion Proceedings of the ACM Web Conference 2025; April 28-May 2, 2025; Sydney, NSW, Australia Companion Proceedings of the ACM Web Conference 2025 (WWW Companion ’25), April 28-May 2, 2025, Sydney, NSW, Australia by Xiaoshuang Chen, Yibo Wang, Yao Wang, Husheng Liu, Kaiqiao Zhan, Ben Wang, Kun Gai (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-20497.txt.

Introduction

Chen and co-authors presented a 2025 paper on recommender systems, the software that picks what users see on video and shopping apps. Most such systems serve the viewer, so a few items get most views while many get almost none. Creators who get little feedback lose interest, which harms the platform over time. The team instead builds for the creator side.

Their system, called DualRec, finds the best users for each item rather than the best items for each user. Known viewer-side methods can be adapted with small changes, they show. A new module also checks whether a user is likely to be active. The system runs on the Kwai video app and improved results there.

Catalog record

CREATORXiaoshuang Chen, Yibo Wang, Yao Wang, Husheng Liu, Kaiqiao Zhan, Ben Wang, Kun Gai
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-20497

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Creator-Side Recommender System: Challenges, Designs, and Applications Conference: Companion Proceedings of the ACM Web Conference 2025; April 28-May 2, 2025; Sydney, NSW, Australia Companion Proceedings of the ACM Web Conference 2025 (WWW Companion ’25), April 28-May 2, 2025, Sydney, NSW, Australia DOI: 10.1145/3701716.3715215 ISBN: 979-8-4007-1331-6/2025/04 CCS: Information systems Recommender systems Xiaoshuang Chen, Yibo Wang, Yao Wang, Husheng Liu, Kaiqiao Zhan, Ben Wang, Kun Gai Abstract . Users and creators are two crucial components of recommender systems. Typical recommender systems focus on the user side, providing the most suitable items based on each user’s request. In such scenarios, a few items receive a majority of exposures, while many items receive very few. This imbalance leads to poorer experiences and decreased activity among the creators receiving less feedback, harming the recommender system in the long term. To this end, we develop a creator-side recommender system, called DualRec , to answer the following question: How to find the most suitable users for each item to enhance the creators’ experience? We show that typical user-side recommendation algorithms, such as retrieval and ranking algorithms, can be adapted into the creator-side versions with just a few modifications. This greatly simplifies algorithm design in DualRec. Moreover, we discuss a unique challenge in DualRec: the user availability issue, which is not present in user-side recommender systems. To tackle this issue, we incorporate a user availability calculation (UAC) module to effectively enhance DualRec’s performance. DualRec has already been implemented in Kwai, a short video recommendation system with over 100 million users and over 10 million creators, significantly improving the experience for creators. 1. Introduction Table 1. Comparisons between user-side and creator-side recommender systems. User-Side Recommender System Creator-Side Recommender System (DualRec) Request Source User u\in\mathcal{U} Item i\in\mathcal{C} Candidate Set the item set \mathcal{C} the user set \mathcal{U} Results an item set \mathcal{C}_{u} to be recommended to User u a user set \mathcal{U}_{i} who will receive Item i Objective user experience on the recommended items creator experience when receiving feedback from the users Figure 1. Impacts of Engagement on Future Post Probability. Figure 2. The DualRec Framework. Recommender systems are widely employed in areas such as short video platforms ( Tang et al. 2017 ; Wu et al. 2018 ) , search engines ( Huang et al. 2020 ; Si et al. 2023 ; Shi et al. 2024 ) , and E-commerce ( Gu et al. 2020 ; Linden et al. 00 3 ; Zhou et al. 2018 ; Xiao et al. 2024 ) to connect users with relevant content. Typically, these systems aim to provide the most relevant items for each user request to enhance user satisfaction. We refer to these systems as user-side recommender systems . User-side recommendations often result in a limited number of items receiving significant exposure, while many creators and their content remain unseen by consumers. Receiving fewer exposures and less feedback typically results in poorer experiences and decreased activity among creators. Recently, there has been a growing body of literature on dispatching long-tail items, such as Google’s cold start system ( Wang et al. 2023 ) , FairRec ( Patro et al. 2020 ) , and others ( Zhang et al. 2021 ; Zhu et al. 2021 ; Zhang et al. 2023 ; Yuan et al. 2023 ) . These approaches create more opportunities for underexposed items, but they still prioritize user satisfaction when distributing these underexposed items. In a real-world recommender system, users and content creators respond differently to various types of feedback. For instance, in Kwai, a short video platform, a user’s watch time on a video is crucial for reflecting user satisfaction; however, this watch time is not visible to the video’s creator and, therefore, cannot impact the creator’s satisfaction. In contrast, creators tend to feel more satisfied when they receive likes and comments on their videos. As shown in Figure 1 , interactions such as likes and comments received by creators have a significant positive incentive effect on their content creation. This disparity between user and creator satisfaction presents challenges for recommender systems to consider the creator experience. To this end, this paper discusses the development of a creator-side recommender system aimed at finding the most appropriate users for each item to enhance creator satisfaction . This creator-side recommender system is called DualRec , where “dual” signifies its correspondence with user-side recommender systems. DualRec is a mirror system to user-side recommender systems, with all concepts from the user-side recommender system having corresponding concepts in DualRec. Table 1 presents a comparison between user-side recommender systems and DualRec. We first demonstrate that there are many similarities between traditional user-side recommender systems and DualRec. Specifically, a typical user-side recommender system, illustrated in Figure 2 (a), consists of multiple retrieval modules that generate candidate item sets, along with a ranking module that predicts user feedback on the items for final recommendations. Similarly, DualRec, as shown in Figure 2 (b), also includes retrieval and ranking modules. Furthermore, we illustrate that typical algorithms used in traditional user-side recommender systems can be adapted into their corresponding creator-side versions with just a few modifications. This feature ensures the compatibility of DualRec with user-side recommendation systems, allowing research related to the latter to be directly applied to the creator-side recommendation system. We then demonstrate that, unlike user-side recommender systems, creator-side recommender systems face the user availability issue . In user-side recommender systems, when we want to recommend an item for a user request, we can generally find the item as long as it hasn’t been deleted by the author. However, in creator-side recommendation systems, when we seek to identify suitable users for an item, there is no guarantee that the users will receive it. On one hand, users may simply not visit the platform. For instance, the number of daily active users is about 30% of the total user base on the Kwai platform. On the other hand, there are no constraints on the number of items assigned to the same user, and this allows for many items to be allocated to one user, which contrasts with the reality that a single user cannot consume too many items. To address this issue, we introduce a user availability calculation(UAC) module to calculate and maintain available users and use UAC to guide the creator-side recommendations, as shown in Figure 2 (c). In summary, our contributions are as follows: • We introduce a novel creator-side recommender system called DualRec, designed to find the most appropriate users for each item to enhance creator satisfaction, which cannot be covered by traditional user-side recommender systems. • We discuss the detailed implementation of DualRec, and show that existing user-side recommendation algorithms can be mirrored to the creator-side version with a few changes, which simplifies the implementation of DualRec. • We address the user availability issue, a unique challenge in creator-side recommendation systems, and introduce UAC to overcome this challenge. • We implement DualRec on the Kwai platform, showing that DualRec significantly enhances the creator experience. 2. Problem Formulation Before introducing the creator-side recommendation, we first formally discuss the user-side recommendation. Denote the user set as \mathcal{U} and the item set as \mathcal{C} . When a user u\in\mathcal{U} sends a request to the system, the user-side recommender system

Read the complete text