Modern applications provided personalized recommendations across diverse scenarios, including the homepage, local pages, and live streams on platforms like TikTok. These scenarios exhibit varying user behavior patterns, resulting in heterogeneous yet interrelated distributions. Existing Multi-Scenario Recommendation (MSR) methods usually use parameter-sharing networks for shared features and scenario-specific networks for unique features. However, these methods fail to handle different distribution across scenarios, resulting in representation entanglement and localization, which hinder effective knowledge transfer and compromise performance. In this paper, we propose a Distribution-aware Re-representations (DAR) method for MSR. Its core idea is to construct distribution-aware prototype spaces and learn disentangled re-representations around global prototypes. Specifically, DAR employs a Multi-gate Mixture of Experts (MMoE) to obtain scenario-shared representations, and uses independent networks to learn scenario-specific representations. These representations are then projected into scenario-shared and scenario-specific prototype spaces, producing scenario-shared re-representations (capturing global information) and scenario-specific re-representations (focusing on distributional differences). During this process, DAR utilizes Unbalanced Optimal Transport (UOT) to compute the transport relationships between representations and global prototypes, taking these as pseudo-labels for re-representation learning. Moreover, to prevent prototype entanglement, a matrix orthogonalization constraint ensures independence among global prototypes. The effectiveness of DAR is demonstrated through extensive offline experiments conducted on four datasets, as well as online A/B tests on a video platform.

Meme-based social abuse detection is challenging because harmful intent often relies on implicit cultural symbolism and subtle cross-modal incongruence. Prior approaches, from fusion-based methods to in-context learning with Large Vision-Language Models (LVLMs), have made progress but remain limited by three factors: i) cultural blindness (missing symbolic context), ii) boundary ambiguity (satire vs. abuse confusion), and iii) lack of interpretability (opaque model reasoning). We introduce CROSS-ALIGN+, a three-stage framework that systematically addresses these limitations: (1) Stage I mitigates cultural blindness by enriching multimodal representations with structured knowledge from ConceptNet, Wikidata, and Hatebase; (2) Stage II reduces boundary ambiguity through parameter-efficient LoRA adapters that sharpen decision boundaries; and (3) Stage III enhances interpretability by generating cascaded explanations. Extensive experiments on five benchmarks and eight LVLMs demonstrate that CROSS-ALIGN+ consistently outperforms state-of-the-art methods, achieving up to 17% relative F1 improvement while providing interpretable justifications for each decision.

Large-scale recommendation systems mainly rely on sparse ID features, struggling with data sparsity. It's important to use multimodal information to assist ID learning for better performance. However, there exists two challenges: (1) distribution discrepancy between multimodal and ID makes direct integration prone to user-item mismatch; (2) slower convergence of multimodal representations compared to ID, causing optimization imbalance under a unified objective, which limits the potential of multimodal representations. In this paper, we comprehensively investigate the two problems and proposes a framework named AB-Rec to align and balance ID and multimodal representations learning for recommendation. We design three alignment tasks to fine-tune a pre-trained multimodal large language model (MLLM), which is then utilized to generate a unified multimodal representation for each item. AB-Rec aligns the distributions of ID and multimodal representations by minimizing the in-batch Wasserstein distance, and maximizes the distance between the two types of representations for the same item to avoid representation collapse. To solve the optimization imbalance, we propose a gradient modulation method that adaptively controls the optimization process by monitoring the contribution differences between ID and multimodal representations. Finally, we conduct extensive offline experiments on four datasets and an A/B test on an online video platform, demonstrating the effectiveness and scalability of our proposed method.

In video recommendation systems, accurately predicting watch time is crucial for enhancing user engagement and retention. Traditional methods typically apply label transformations or mitigate duration bias to improve performance but overlook that erroneous instance representations are the primary cause of significant prediction errors. Moreover, these approaches predominantly rely on point perdition, limiting their robustness. To address these challenges, we propose LDA, a novel prediction paradigm that optimizes instance representations by explicitly leveraging label distributions as anchors within the model, enabling more accurate and robust predictions. Our analysis reveals that watch ratio across different duration groups exhibit distinct multi-peak distributions, reflecting the strong aggregation of user behavior. Based on this finding, we employ Vector Quantized Variational Auto-encoder (VQ-VAE) to convert the continuous watch ratio distribution into representative anchors that capture these multi-peak characteristics within each duration group. Subsequently, we project both instance representations and anchors into a common space and utilize Optimal Transport (OT) to generate pseudo-labels aligned with the anchor distribution, allowing instances to obtain structured coordinates within this space during training. Finally, we derive optimized instance representations for watch time prediction by aggregating anchor vectors through weighted integration. Extensive offline experiments on two datasets and large-scale online A/B testing on a short-video platform with over 300 million DAUs demonstrate the consistent superiority of LDA in watch time prediction.

In Internet of Vehicles (IoV), intelligent transportation recognition is key to smart transportation systems. However, training models using data from individual vehicles often results in unreliable performance due to limited storage, leading to new data overwriting old samples. This approach suffers from catastrophic forgetting, low accuracy, and poor generalization. Current industry solutions, such as federated learning, incremental learning (iCaRL), and elastic weight consolidation (EWC), face limitations when dealing with the frequent, dynamic changes in IoV data and the heterogeneity of system structures.In this paper, we propose the FedHotpot framework based on federated learning to tackle these challenges. On the one hand, it employs a dual knowledge distillation strategy to ensure that knowledge learned by the iteratively updated model is not forgotten. On the other hand, we have optimized the model aggregation algorithm to mitigate the adverse effects of system and statistical heterogeneity arising from the diverse clients in the IoV environment on the federated learning architecture. We conducted multiple sets of comparative experiments on the proposed FedHotpot framework. The experimental results demonstrate that FedHotpot performs exceptionally well across multiple datasets, particularly when dealing with non-independently and identically distributed (non-IID) data, where its performance significantly outperforms traditional federated learning methods. These findings not only validate the effectiveness of the dual knowledge distillation strategy, but also underscore the pivotal role of optimized aggregation algorithms in enhancing the overall performance of federated learning.

A large-scale cloud data center needs to provide high service reliability and availability with low failure occurrence probability. However, current large-scale cloud data centers still face high failure rates due to many reasons such as hardware and software failures, which often result in task and job failures. Such failures can severely reduce the reliability of cloud services and also occupy huge amount of resources to recover the service from failures. Therefore, it is important to predict task or job failures before occurrence with high accuracy to avoid unexpected wastage. Many machine learning and deep learning based methods have been proposed for the task or job failure prediction by analyzing past system message logs and identifying the relationship between the data and the failures. In order to further improve the failure prediction accuracy of the previous machine learning and deep learning based methods, in this paper, we propose a failure prediction algorithm based on multi-layer Bidirectional Long Short Term Memory (Bi-LSTM) to identify task and job failures in the cloud. The goal of Bi-LSTM prediction algorithm is to predict whether the tasks and jobs are failed or completed. The trace-driven experiments show that our algorithm outperforms other state-of-art prediction methods with 93% accuracy and 87% for task failure and job failures respectively.