A5086177569- Recommender Systems and Techniques
- Advanced Bandit Algorithms Research
- Image and Video Quality Assessment
- Reinforcement Learning in Robotics
- Human Mobility and Location-Based Analysis
- Smart Grid Energy Management
- Caching and Content Delivery
- Video Analysis and Summarization
- Data Stream Mining Techniques
- Multimedia Communication and Technology
- Mobile Crowdsensing and Crowdsourcing
Kuaishou (China)
2022-2023
Watch-time prediction remains to be a key factor in reinforcing user engagement via video recommendations. It has become increasingly important given the ever-growing popularity of online videos. However, watch time not only depends on match between and but is often mislead by duration itself. With goal improving time, recommendation always biased towards videos with long duration. Models trained this imbalanced data face risk bias amplification, which misguides platforms over-recommend...
In recent years, Multi-task Learning (MTL) has yielded immense success in Recommender System (RS) applications [40]. However, current MTL-based recommendation models tend to disregard the session-wise patterns of user-item interactions because they are predominantly constructed based on item-wise datasets. Moreover, balancing multiple objectives always been a challenge this field, which is typically avoided via linear estimations existing works. To address these issues, paper, we propose...
The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems the video-sharing platforms. Users sequentially interact with system provide complex multi-faceted responses, including WatchTime various types interactions multiple videos. On one hand, platforms aim at optimizing users' cumulative (main goal) in long term, which can be effectively optimized by Reinforcement Learning. other also need satisfy constraint accommodating...
Recently, short video platforms have achieved rapid user growth by recommending interesting content to users. The objective of the recommendation is optimize retention, thereby driving DAU (Daily Active Users). Retention a long-term feedback after multiple interactions users and system, it hard decompose retention reward each item or list items. Thus traditional point-wise list-wise models are not able retention. In this paper, we choose reinforcement learning methods as they designed...
In recommender systems, reinforcement learning solutions have effectively boosted recommendation performance because of their ability to capture long-term user-system interaction. However, the action space policy is a list items, which could be extremely large with dynamic candidate item pool. To overcome this challenge, we propose hyper-actor and critic framework where decomposes generation process into hyper-action inference step an effect-action selection step. The first maps given state...
Current advances in recommender systems have been remarkably successful optimizing immediate engagement. However, long-term user engagement, a more desirable performance metric, remains difficult to improve. Meanwhile, recent reinforcement learning (RL) algorithms shown their effectiveness variety of goal optimization tasks. For this reason, RL is widely considered as promising framework for engagement recommendation. Though promising, the application heavily relies on well-designed rewards,...
Personalized recommender systems fulfill the daily demands of customers and boost online businesses. The goal is to learn a policy that can generate list items matches user's demand or interest. While most existing methods pointwise scoring model predicts ranking score each individual item, recent research shows listwise approach further improve recommendation quality by modeling intra-list correlations are exposed together. This has motivated reranking generative approaches optimize overall...
Reinforcement Learning (RL)-based recommender systems (RSs) have garnered considerable attention due to their ability learn optimal recommendation policies and maximize long-term user rewards. However, deploying RL models directly in online environments generating authentic data through A/B tests can pose challenges require substantial resources. Simulators offer an alternative approach by providing training evaluation for RS models, reducing reliance on real-world data. Existing simulators...
With the ever prospering of web technologies, there is a common need to make recommendations from heterogeneous sources, such as recommending products and advertisements together on e-commerce websites. People usually solve recommendation problem by two-stage paradigm, where first stage generating candidates each source, second one aggregating ranking generated produce final results. While existing models have achieved many successes, they mostly optimize above two stages separately, user...
Real-world recommendation systems often consist of two phases. In the first phase, multiple predictive models produce probability different immediate user actions. second these predictions are aggregated according to a set 'strategic parameters' meet diverse business goals, such as longer engagement, higher revenue potential, or more community/network interactions. addition building accurate models, it is also crucial optimize this so that primary goals optimized while secondary guardrails...