xMTF: A Formula-Free Model for Reinforcement-Learning-Based Multi-Task Fusion in Recommender Systems

Recommender systems need to optimize various types of user feedback, e.g., clicks, likes, and shares. A typical recommender system handling multiple types of feedback has two components: a multi-task learning (MTL) module, predicting feedback such as click-through rate and like rate; and a multi-task fusion (MTF) module, integrating these predictions into a single score for item ranking. MTF is essential for ensuring user satisfaction, as it directly influences recommendation outcomes. Recently, reinforcement learning (RL) has been applied to MTF tasks to improve long-term user satisfaction. However, existing RL-based MTF methods are formula-based methods, which only adjust limited coefficients within pre-defined formulas. The pre-defined formulas restrict the RL search space and become a bottleneck for MTF. To overcome this, we propose a formula-free MTF framework. We demonstrate that any suitable fusion function can be expressed as a composition of single-variable monotonic functions, as per the Sprecher Representation Theorem. Leveraging this, we introduce a novel learnable monotonic fusion cell (MFC) to replace pre-defined formulas. We call this new MFC-based model eXtreme MTF (xMTF). Furthermore, we employ a two-stage hybrid (TSH) learning strategy to train xMTF effectively. By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments.

Unleashing the Potential of Two-Tower Models: Diffusion-Based Cross-Interaction for Large-Scale Matching

Two-tower models are widely adopted in the industrial-scale matching stage across a broad range of application domains, such as content recommendations, advertisement systems, and search engines. This model efficiently handles large-scale candidate item screening by separating user and item representations. However, the decoupling network also leads to a neglect of potential information interaction between the user and item representations. Current state-of-the-art (SOTA) approaches include adding a shallow fully connected layer(i.e., COLD), which is limited by performance and can only be used in the ranking stage. For performance considerations, another approach attempts to capture historical positive interaction information from the other tower by regarding them as the input features(i.e., DAT). Later research showed that the gains achieved by this method are still limited because of lacking the guidance on the next user intent. To address the aforementioned challenges, we propose a "cross-interaction decoupling architecture" within our matching paradigm. This user-tower architecture leverages a diffusion module to reconstruct the next positive intention representation and employs a mixed-attention module to facilitate comprehensive cross-interaction. During the next positive intention generation, we further enhance the accuracy of its reconstruction by explicitly extracting the temporal drift within user behavior sequences. Experiments on two real-world datasets and one industrial dataset demonstrate that our method outperforms the SOTA two-tower models significantly, and our diffusion approach outperforms other generative models in reconstructing item representations.

Creator-Side Recommender System: Challenges, Designs, and Applications

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 millions user and over 10 million creators, significantly improving the experience for creators.

MARM: Unlocking the Future of Recommendation Systems through Memory Augmentation and Scalable Complexity

Scaling-law has guided the language model designing for past years, however, it is worth noting that the scaling laws of NLP cannot be directly applied to RecSys due to the following reasons: (1) The amount of training samples and model parameters is typically not the bottleneck for the model. Our recommendation system can generate over 50 billion user samples daily, and such a massive amount of training data can easily allow our model parameters to exceed 200 billion, surpassing many LLMs (about 100B). (2) To ensure the stability and robustness of the recommendation system, it is essential to control computational complexity FLOPs carefully. Considering the above differences with LLM, we can draw a conclusion that: for a RecSys model, compared to model parameters, the computational complexity FLOPs is a more expensive factor that requires careful control. In this paper, we propose our milestone work, MARM (Memory Augmented Recommendation Model), which explores a new cache scaling-laws successfully.

GPT-4o System Card

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

Enhancing Playback Performance in Video Recommender Systems with an On-Device Gating and Ranking Framework

Video recommender systems (RSs) have gained increasing attention in recent years. Existing mainstream RSs focus on optimizing the matching function between users and items. However, we noticed that users frequently encounter playback issues such as slow loading or stuttering while browsing the videos, especially in weak network conditions, which will lead to a subpar browsing experience, and may cause users to leave, even when the video content and recommendations are superior. It is quite a serious issue, yet easily overlooked. To tackle this issue, we propose an on-device Gating and Ranking Framework (GRF) that cooperates with server-side RS. Specifically, we utilize a gate model to identify videos that may have playback issues in real-time, and then we employ a ranking model to select the optimal result from a locally-cached pool to replace the stuttering videos. Our solution has been fully deployed on Kwai, a large-scale short video platform with hundreds of millions of users globally. Moreover, it significantly enhances video playback performance and improves overall user experience and retention rates.

Cache-Aware Reinforcement Learning in Large-Scale Recommender Systems

Modern large-scale recommender systems are built upon computation-intensive infrastructure and usually suffer from a huge difference in traffic between peak and off-peak periods. In peak periods, it is challenging to perform real-time computation for each request due to the limited budget of computational resources. The recommendation with a cache is a solution to this problem, where a user-wise result cache is used to provide recommendations when the recommender system cannot afford a real-time computation. However, the cached recommendations are usually suboptimal compared to real-time computation, and it is challenging to determine the items in the cache for each user. In this paper, we provide a cache-aware reinforcement learning (CARL) method to jointly optimize the recommendation by real-time computation and by the cache. We formulate the problem as a Markov decision process with user states and a cache state, where the cache state represents whether the recommender system performs recommendations by real-time computation or by the cache. The computational load of the recommender system determines the cache state. We perform reinforcement learning based on such a model to improve user engagement over multiple requests. Moreover, we show that the cache will introduce a challenge called critic dependency, which deteriorates the performance of reinforcement learning. To tackle this challenge, we propose an eigenfunction learning (EL) method to learn independent critics for CARL. Experiments show that CARL can significantly improve the users' engagement when considering the result cache. CARL has been fully launched in Kwai app, serving over 100 million users.

Cognitively Biased Users Interacting with Algorithmically Biased Results in Whole-Session Search on Controversial Topics

When interacting with information retrieval (IR) systems, users, affected by confirmation biases, tend to select search results that confirm their existing beliefs on socially significant contentious issues. To understand the judgments and attitude changes of users searching online, our study examined how cognitively biased users interact with algorithmically biased search engine result pages (SERPs). We designed three-query search sessions on debated topics under various bias conditions. We recruited 1,321 crowdsourcing participants and explored their attitude changes, search interactions, and the effects of confirmation bias. Three key findings emerged: 1) most attitude changes occur in the initial query of a search session; 2) confirmation bias and result presentation on SERPs affect search behaviors in the current query and perceived familiarity with clicked results in subsequent queries. The bias position also affect attitude changes of users with lower perceived openness to conflicting opinions; 3) Interactions in the first query and and dwell time throughout the session are associated with users' attitude changes in different forms. Our study goes beyond traditional simulation-based evaluation settings and simulated rational users, sheds light on the mixed effects of human biases and algorithmic biases in controversial information retrieval tasks, and can inform the design of bias-aware user models, human-centered bias mitigation techniques, and socially responsible intelligent IR systems.

GOLF: Goal-Oriented Long-term liFe tasks supported by human-AI collaboration

The advent of ChatGPT and similar large language models (LLMs) has revolutionized the human-AI interaction and information-seeking process. Leveraging LLMs as an alternative to search engines, users can now access summarized information tailored to their queries, significantly reducing the cognitive load associated with navigating vast information resources. This shift underscores the potential of LLMs in redefining information access paradigms. Drawing on the foundation of task-focused information retrieval and LLMs' task planning ability, this research extends the scope of LLM capabilities beyond routine task automation to support users in navigating long-term and significant life tasks. It introduces the GOLF framework (Goal-Oriented Long-term liFe tasks), which focuses on enhancing LLMs' ability to assist in significant life decisions through goal orientation and long-term planning. The methodology encompasses a comprehensive simulation study to test the framework's efficacy, followed by model and human evaluations to develop a dataset benchmark for long-term life tasks, and experiments across different models and settings. By shifting the focus from short-term tasks to the broader spectrum of long-term life goals, this research underscores the transformative potential of LLMs in enhancing human decision-making processes and task management, marking a significant step forward in the evolution of human-AI collaboration.

Seed Optimization with Frozen Generator for Superior Zero-shot Low-light Enhancement

In this work, we observe that the generators, which are pre-trained on massive natural images, inherently hold the promising potential for superior low-light image enhancement against varying scenarios.Specifically, we embed a pre-trained generator to Retinex model to produce reflectance maps with enhanced detail and vividness, thereby recovering features degraded by low-light conditions.Taking one step further, we introduce a novel optimization strategy, which backpropagates the gradients to the input seeds rather than the parameters of the low-light enhancement model, thus intactly retaining the generative knowledge learned from natural images and achieving faster convergence speed. Benefiting from the pre-trained knowledge and seed-optimization strategy, the low-light enhancement model can significantly regularize the realness and fidelity of the enhanced result, thus rapidly generating high-quality images without training on any low-light dataset. Extensive experiments on various benchmarks demonstrate the superiority of the proposed method over numerous state-of-the-art methods qualitatively and quantitatively.