Outlier detection (OD) aims to identify abnormal instances, known as outliers or anomalies, by learning typical patterns of normal data, or inliers. Performing OD under an unsupervised regime--without any information about anomalous instances in the training data--is challenging. A recently observed phenomenon, known as the inlier-memorization (IM) effect, where deep generative models (DGMs) tend to memorize inlier patterns during early training, provides a promising signal for distinguishing outliers. However, existing unsupervised approaches that rely solely on the IM effect still struggle when inliers and outliers are not well-separated or when outliers form dense clusters. To address these limitations, we incorporate active learning to selectively acquire informative labels, and propose IMBoost, a novel framework that explicitly reinforces the IM effect to improve outlier detection. Our method consists of two stages: 1) a warm-up phase that induces and promotes the IM effect, and 2) a polarization phase in which actively queried samples are used to maximize the discrepancy between inlier and outlier scores. In particular, we propose a novel query strategy and tailored loss function in the polarization phase to effectively identify informative samples and fully leverage the limited labeling budget. We provide a theoretical analysis showing that the IMBoost consistently decreases inlier risk while increasing outlier risk throughout training, thereby amplifying their separation. Extensive experiments on diverse benchmark datasets demonstrate that IMBoost not only significantly outperforms state-of-the-art active OD methods but also requires substantially less computational cost.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
Graph Contrastive Learning (GCL) has recently emerged as a powerful paradigm for modeling user–item interactions and learning high-quality representations in recommender systems. While existing GCL-based methods benefit from data augmentation and sampling strategies, they often overlook the inherent limitations of the contrastive objectives: 1) Stacking multiple Graph Convolutional Network layers to capture high-order information often causes the over-smoothing phenomenon, where node representations become overly similar. 2) Structurally similar negative sample pairs may exhibit high cosine similarity, causing gradient saturation during representation optimization. To address the above challenges, we revisit matrix factorization in recommendation models and uncover its implicit connection to a parallel graph filter bank. This perspective reveals how overly aggressive low-pass or high-pass filtering distorts feature distributions, contributing to gradient saturation. Building on this insight, we propose Light Cosine Similarity Collaborative Filtering (LightCSCF), a margin-constrained method that improves gradient optimization in contrastive learning by focusing on structurally hard examples, alleviating both gradient saturation and boundary over-smoothing. Extensive experiments on three real-world datasets demonstrate that LightCSCF consistently outperforms state-of-the-art baselines in recommendation accuracy and robustness to data sparsity.
Diffusion models have demonstrated remarkable success in generative tasks, including audio super-resolution (SR). In many applications like movie post-production and album mastering, substantial computational budgets are available for achieving superior audio quality. However, while existing diffusion approaches typically increase sampling steps to improve quality, the performance remains fundamentally limited by the stochastic nature of the sampling process, leading to high-variance and quality-limited outputs. Here, rather than simply increasing the number of sampling steps, we propose a different paradigm through inference-time scaling for SR, which explores multiple solution trajectories during the sampling process. Different task-specific verifiers are developed, and two search algorithms, including the random search and zero-order search for SR, are introduced. By actively guiding the exploration of the high-dimensional solution space through verifier-algorithm combinations, we enable more robust and higher-quality outputs. Through extensive validation across diverse audio domains (speech, music, sound effects) and frequency ranges, we demonstrate consistent performance gains, achieving improvements of up to 9.70% in aesthetics, 5.88% in speaker similarity, 15.20% in word error rate, and 46.98% in spectral distance for speech SR from 4 kHz to 24 kHz, showcasing the effectiveness of our approach.
Heterogeneous graphs are widely used to model real-world systems with diverse entity types and relational structures, and existing methods have shown promising performance in various applications. However, most current models assume balanced and semantically aligned features across nodes, which rarely holds in practice. In scenarios such as social risk governance, node types often exhibit severe feature imbalance, making it difficult for standard aggregation mechanisms to extract meaningful signals. This imbalance leads to three key challenges: inaccurate neighbor weighting, noise propagation, and biased representations skewed toward text-rich nodes. To address these issues, we propose HeCoGNN, a collaborative and adaptive aggregation framework that jointly performs neighbor filtering and relation-aware message calibration, enabling robust representation learning under semantic disparity. Experiments on real-world social governance graphs show that HeCoGNN consistently outperforms state-of-the-art baselines, particularly in handling underrepresented and noisy node types.
Video recommendation systems heavily rely on user watch time feedback, making accurate watch time prediction a crucial task. However, this task inherently suffers from bias, as recommendation models tend to favor long-duration videos to maximize watch time. This issue, known as duration bias in the watch-time prediction context, can be explained from a causal perspective, where video duration acts as a confounder. Recent works address this bias using backdoor adjustment, isolating the direct effect of content on watch time from observational data. These methods typically discretize video duration into groups, estimate group-wise effects, and then aggregate them via a unified prediction model. However, this aggregation strategy is prone to model misspecification due to feature distribution shift across groups. In this paper, we reinterpret the problem through the lens of invariant learning and propose a novel framework: Duration-Invariant Feature Learning (DIFL). DIFL employs a kernel-based regularization that enforces representation invariance across duration groups, reducing sensitivity to group design and improving generalization. This enables more accurate modeling of the direct causal effect and making counterfactual inference. Extensive experiments on both public and real large-scale production datasets demonstrate the effectiveness of our approach, which achieves SOTA performance.
In recommender systems, recent advances highlight the critical role of alignment and uniformity (AU) in representation learning. Specifically, AU-based methods pull positive user-item pairs closer (alignment) and spread the overall representation distribution (uniformity), typically relying on observed positive samples. Despite their effectiveness, exist methods face two limitations: (1) noise issues have a more severe impact on AU-based methods in the absence of negative samples, leading to the capture of spurious signals such as misclicks or non-preferential behaviors; (2) data sparsity weakens the alignment of user-item representations, hindering reliable representation learning and harming recommendations for sparse users. To tackle these issues, we propose a novel recommendation framework named Constrained Alignment and Filtered Uniformity (CAFU). CAFU enhances robustness through Filtered Uniformity (FU) and improves performance under data sparsity via Constrained Alignment (CA). Specifically, FU adopts a threshold-based strategy to eliminate unreliable samples that degrade embedding quality, thereby strengthening robustness. In parallel, CA mitigates the impact of sparsity by masking low-confidence user-item pairs based on angular distance, leading to better recommendation for sparse users. Extensive experiments on three datasets and three backbones demonstrate the effectiveness and generalization of the proposed framework.
The continuous emergence of new entities, relations, triples, and multimodal information drives the dynamic evolution of multimodal knowledge graph (MMKG). However, existing MMKG embedding models follow a static setting, where training from scratch for growing MMKG wastes learned knowledge, while fine-tuning on new knowledge easily leads to catastrophic forgetting, severely limiting their applicability in real-world scenarios. To address this, we propose a multimodal continual representation learning framework (MoFot) for growing MMKG. Unlike existing static multimodal embedding methods, MoFot focuses on alleviating catastrophic forgetting rather than retraining to adapt to new knowledge. Specifically, MoFot effectively mitigates catastrophic forgetting caused by parameter updates and differing forgetting rates across modalities through a multimodal collaborative modulation mechanism. The mechanism ensures consistent retention of previously learned multimodal knowledge across snapshots through multimodal weight modulation and multimodal feature modulation. MoFot outperforms existing MMKG embedding, KG continual learning, and MMKG inductive models. Experimental results demonstrate that MoFot not only avoids forgetting but also enhances old knowledge by learning new knowledge, achieving adaptation to new knowledge while mitigating forgetting of old knowledge.
Perceptual image compression has recently gained increasing attention, as it aims to reconstruct visually realistic images using generative models. Most existing methods adopt patch-based generative adversarial networks (PatchGAN) for one-step image generation, where adversarial training helps the decoder learn the distribution of natural images. However, this strategy is often coarse-grained, as it focuses mainly on patch-level consistency and overlooks global structural and semantic details. To address this limitation, we propose a simple yet effective Semantic and Spectral Consistency Learning (SSCL) strategy, which complements existing patch-based approaches for more accurate distribution alignment. For semantic consistency, we leverage semantic vision models to extract semantic features. The semantic discriminator, aware of the specific semantics of each image, provides more adaptive and precise feedback. This guides the encoder to retain meaningful information and helps the decoder synthesize detailed textures, without requiring explicit semantic transmission or additional modules. For spectral consistency, we introduce a frequency discriminator that focuses on high-frequency components, helping to reduce artifacts based on spectral priors. Experiments show that SSCL outperforms existing perceptual codecs in terms of visual quality. Compared to MS-ILLM, SSCL achieves 45% to 60% bit-rate savings on CLIC2020 and Kodak datasets, measured by FID and DISTS.
Recent advances in Large Language Models (LLMs) have enhanced text-based recommendation by enriching traditional ID-based methods with semantic generalization capabilities. Text-based methods typically encode item textual information via prompt design and generate discrete semantic IDs through item tokenization. However, in domain-specific tasks such as local-life services, simply injecting location information into prompts fails to capture fine-grained spatial characteristics and real-world distance awareness among items. To address this, we propose LGSID, an LLM-Aligned Geographic Item Tokenization Framework for Local-life Recommendation. This framework consists of two key components: (1) RL-based Geographic LLM Alignment, and (2) Hierarchical Geographic Item Tokenization. In the RL-based alignment module, we initially train a list-wise reward model to capture real-world spatial relationships among items. We then introduce a novel G-DPO algorithm that uses pre-trained reward model to inject generalized spatial knowledge and collaborative signals into LLMs while preserving their semantic understanding. Furthermore, we propose a hierarchical geographic item tokenization strategy, where primary tokens are derived from discrete spatial and content attributes, and residual tokens are refined using the aligned LLM’s geographic representation vectors. Extensive experiments on real-world Kuaishou industry datasets show that LGSID consistently outperforms state-of-the-art discriminative and generative recommendation models. Ablation studies, visualizations, and case studies further validate its effectiveness.
DuoKD: Dual Knowledge Distillation from Large Language Models for Robust Graph Neural Networks
PDF ↗Graph neural networks (GNNs) have become a dominant modeling paradigm for graph-structured data, and the emergence of large language models (LLMs) has spurred growing interest in integrating external semantic knowledge into GNNs. Current LLM-based GNNs are devoted to extracting semantically similar information from LLMs to enhance representation learning. However, they generally overlook key signals that are semantically dissimilar but exhibit stronger inter-class discriminative ability. Especially when the original graph data contains noise or semantic ambiguity, a single similarity-based semantic augmentation strategy not only fails to provide effective enhancement, but may also amplify misleading signals generated by the LLM in response to low-quality inputs or its own hallucinations, further degrading the discriminative power and robustness of GNNs. To this end, we propose a dual positive-negative knowledge extraction strategy based on LLMs, and integrate it with a knowledge distillation mechanism to dynamically transfer multi-dimensional enhanced signals to GNNs, thereby achieving fine-grained and robust graph representation learning. Specifically, we design personalized prompts to guide LLMs in generating semantically similar positive signals and semantically dissimilar negative signals, which help the model capture intra-class consistency and inter-class distinction. Then, we further generate structural and semantic reasoning as supplementary knowledge to support the rationality and guidance of supervision signals. To identify high-confidence transferred knowledge, we introduce a language-based evaluation mechanism to filter low-confidence or hallucinated outputs. Finally, under a unified distillation framework, our method uses both positive and negative knowledge to guide GNN training, achieving adaptive and robust representation learning. Extensive experiments on benchmark datasets verify the superior performance of our approach across various tasks.
Recent years have witnessed the wide adoption of deep learning recommendation models (DLRMs) for many online services. Unlike traditional DNN training, DLRMs leverage massive embeddings to represent sparse features, which are stored in distributed GPUs following the model parallel paradigm. Existing approaches adopt deduplication to eliminate replicated embeddings involved in AltoAll transfers to avoid unnecessary communication. In our practices, we have observed that such a deduplication design exacerbates interconnect inefficiency due to the fragmented embedding transfers with reduced message sizes, hindering the performance of distributed DLRM training. This paper introduces FusedRec, a fused embedding communication and lookup mechanism to tackle the inefficiency due to deduplication. By seeking the opportunities to fuse embeddings from multiple categories into a group, FusedRec conducts the communication in a combined shot to alleviate bandwidth under-utilization. Meanwhile, a categorical-aware hashing algorithm is integrated into FusedRec to retain the category information during lookup without extra communication. Combining with efficient unique and recovery operations, comprehensive results show FusedRec achieves a 37.8% throughput speedup in average compared to the SOTA industry implementation, without hurting the recommendation qualities of our in-house models used in online production environments.
Developing a universal graph model capable of generalizing across diverse graph domains has consistently been a key objective in graph learning. Recently, many studies have focused on achieving in-context learning (ICL) on graphs, which can generalize to novel tasks without the need for fine-tuning, similar to large language models (LLMs) such as GPT-3. These researches can be primarily divided into graph-based methods and LLM-based methods. However, the generalization performance of the former is limited by the representation capability of GNNs, while the latter faces the challenge of LLMs understanding graph structures. Therefore, we propose CAGML, a context-aware graph meta-learning model, which learns to generalize to cross-domain and cross-granularity graph tasks using a meta-trained Transformer. Firstly, we formulate graph few-shot learning tasks as a structure-aware sequence modeling problem to unify cross-domain and cross-granularity tasks. Then, a structure-aware Transformer (SAT) is introduced as a graph in-context learner to make predictions with a few labels and the task-specific structural context. Finally, we pre-train SAT in a meta-optimization manner on large-scale citation network and knowledge graph. Experiments on 6 cross-domain graph datasets show that, without fine-tuning, CAGML can achieve state-of-the-art (SOTA) performance in terms of average performance across cross-granularity tasks on adopted datasets.
Anomaly detection in dynamic graphs is a critical area of research that focuses on identifying abnormal components within evolving graph structures that deviate significantly from typical patterns. Despite advancements in traditional temporal pattern mining and deep learning techniques, a comprehensive benchmarking framework for Dynamic Graph Anomaly Detection (DyGAD) has been lacking. To address this gap, we introduce BAG, the first comprehensive benchmark specifically designed for anomaly detection on dynamic graphs. BAG enables extensive evaluation of 25 leading DyGAD models, covering both classical approaches and advanced Dynamic Graph Neural Networks (DGNNs), across 10 diverse real-world datasets that include both synthetic and naturally occurring anomalies. The framework supports evaluations at both the edge and node levels, offering a robust tool to advance DyGAD research. Our main finding is that Continuous-time Dynamic Graph (CTDG) models demonstrate superior performance and potential in detecting anomalies in dynamic graph edges, compared to Discrete-time Dynamic Graph (DTDG) models. Furthermore, the results reveal that existing methods are less effective at detecting organic anomalies, primarily due to the presence of temporal anomalies and highly imbalanced samples. The proposed BAG benchmark significantly enhances the evaluation of DyGAD methods by improving dataset selection, metric application, and model training. Moreover, BAG supports reproducibility and further exploration in this field by integrating all models, datasets, and evaluation protocols into an open-source repository.
Stickers are widely used in online communication to convey emotions and implicit intentions. The Sticker Response Selection (SRS) task aims to select the most contextually appropriate sticker based on the dialogue. However, existing methods typically rely on semantic matching and model emotional and intentional cues separately, which can lead to mismatches when emotions and intentions are misaligned. To address this issue, we propose Emotion and Intention Guided Multi-Modal Learning (EIGML). This framework is the first to jointly model emotion and intention, effectively reducing the bias caused by isolated modeling and significantly improving selection accuracy. Specifically, we introduce Dual-Level Contrastive Framework to perform both intra-modality and inter-modality alignment, ensuring consistent representation of emotional and intentional features within and across modalities. In addition, we design an Intention-Emotion Guided Multi-Modal Fusion module that integrates emotional and intentional information progressively through three components: Emotion-Guided Intention Knowledge Selection, Intention-Emotion Guided Attention Fusion, and Similarity-Adjusted Matching Mechanism. This design injects rich, effective information into the model and enables a deeper understanding of the dialogue, ultimately enhancing sticker selection performance. Experimental results on two public datasets show that EIGML outperforms state-of-the-art baselines, achieving higher accuracy and a better understanding of emotional and intentional features.
From IDs to Semantics: A Generative Framework for Cross-Domain Recommendation with Adaptive Semantic Tokenization
PDF ↗Cross-domain recommendation (CDR) is crucial for improving recommendation accuracy and generalization, yet traditional methods are often hindered by the reliance on shared user/item IDs, which are unavailable in most real-world scenarios. Consequently, many efforts have focused on learning disentangled representations through multi-domain joint training to bridge the domain gaps. Recent Large Language Model (LLM)-based approaches show promise, they still face critical challenges, including: (1) the \textbf{item ID tokenization dilemma}, which leads to vocabulary explosion and fails to capture high-order collaborative knowledge; and (2) \textbf{insufficient domain-specific modeling} for the complex evolution of user interests and item semantics. To address these limitations, we propose \textbf{GenCDR}, a novel \textbf{Gen}erative \textbf{C}ross-\textbf{D}omain \textbf{R}ecommendation framework. GenCDR first employs a \textbf{Domain-adaptive Tokenization} module, which generates disentangled semantic IDs for items by dynamically routing between a universal encoder and domain-specific adapters. Symmetrically, a \textbf{Cross-domain Autoregressive Recommendation} module models user preferences by fusing universal and domain-specific interests. Finally, a \textbf{Domain-aware Prefix-tree} enables efficient and accurate generation. Extensive experiments on multiple real-world datasets demonstrate that GenCDR significantly outperforms state-of-the-art baselines. Our code is available in the supplementary materials.
Heterogeneous Graph Neural Networks (HGNNs) are widely used for deep learning on heterogeneous graphs. Typical end-to-end HGNNs require repetitive message passing during training, limiting efficiency for large-scale real-world graphs. Pre-computation-based HGNNs address this by performing message passing only once during preprocessing, collecting neighbor information into regular-shaped tensors, which enables efficient mini-batch training. Label-based pre-computation methods collect neighbors' label information but suffer from training label leakage, where a node's own label information propagates back to itself during multi-hop message passing—the echo effect. Existing mitigation strategies are memory-inefficient on large graphs or suffer from compatibility issues with advanced message passing methods. We propose Echoless Label-based Pre-computation (Echoless-LP), which eliminates training label leakage with Partition-Focused Echoless Propagation (PFEP). PFEP partitions target nodes and performs echoless propagation, where nodes in each partition collect label information only from neighbors in other partitions, avoiding echo while remaining memory-efficient and compatible with any message passing method. We also introduce an Asymmetric Partitioning Scheme (APS) and a PostAdjust mechanism to address information loss from partitioning and distributional shifts across partitions. Experiments on public datasets demonstrate that Echoless-LP achieves superior performance and maintains memory efficiency compared to baselines.
NTSFormer: A Self-Teaching Graph Transformer for Multimodal Isolated Cold-Start Node Classification
PDF ↗Isolated cold-start node classification on multimodal graphs is challenging because such nodes have no edges and often have missing modalities (e.g., absent text or image features). Existing methods address structural isolation by degrading graph learning models to multilayer perceptrons (MLPs) for isolated cold-start inference, using a teacher model (with graph access) to guide the MLP. However, this results in limited model capacity in the student, which is further challenged when modalities are missing. In this paper, we propose Neighbor-to-Self Graph Transformer (NTSFormer), a unified Graph Transformer framework that jointly tackles the isolation and missing-modality issues via a self-teaching paradigm. Specifically, NTSFormer uses a cold-start attention mask to simultaneously make two predictions for each node: a "student" prediction based only on self information (i.e., the node's own features), and a "teacher" prediction incorporating both self and neighbor information. This enables the model to supervise itself without degrading to an MLP, thereby fully leveraging the Transformer's capacity to handle missing modalities. To handle diverse graph information and missing modalities, NTSFormer performs a one-time multimodal graph pre-computation that converts structural and feature data into token sequences, which are then processed by Mixture-of-Experts (MoE) Input Projection and Transformer layers for effective fusion. Experiments on public datasets show that NTSFormer achieves superior performance for multimodal isolated cold-start node classification.
Tokenize Once, Recommend Anywhere: Unified Item Tokenization for Multi-domain LLM-based Recommendation
PDF ↗Large language model (LLM)-based recommender systems have achieved high-quality performance by bridging the discrepancy between the item space and the language space through item tokenization. However, existing item tokenization methods typically require training separate models for each item domain, limiting generalization. Moreover, the diverse distributions and semantics across item domains make it difficult to construct a unified tokenization that preserves domain-specific information. To address these challenges, we propose UniTok, a Unified item Tokenization framework that integrates our own mixture-of-experts (MoE) architecture with a series of codebooks to convert items into discrete tokens, enabling scalable tokenization while preserving semantic information across multiple item domains. Specifically, items from different domains are first projected into a unified latent space through a shared encoder. They are then routed to domain-specific experts to capture the unique semantics, while a shared expert, which is always active, encodes common knowledge transferable across domains. Additionally, to mitigate semantic imbalance across domains, we present a mutual information calibration mechanism, which guides the model towards retaining similar levels of semantic information for each domain. Comprehensive experiments on wide-ranging real-world datasets demonstrate that the proposed UniTok framework is (a) highly effective: achieving up to 51.89% improvements over strong benchmarks, (b) theoretically sound: showing the analytical validity of our architectural design and optimization; and (c) highly generalizable: demonstrating robust performance across diverse domains without requiring per-domain retraining, a capability not supported by existing baselines.
Large Language Models (LLMs) have achieved remarkable success in recent years, owing to their impressive generalization capabilities and rich world knowledge. To capitalize on the potential of using LLMs as recommender systems, mainstream approaches typically focus on two paradigms. The first paradigm designs multi-domain or multi-task instruction data for generalizable recommendation, so as to align LLMs with general recommendation areas and deal with cold-start recommendation. The second paradigm focuses on enhancing domain-specific recommendation tasks, improving performance in warm recommendation scenarios. While most previous works treat these two paradigms separately, we argue that they have complementary advantages, and combining them can yield better results. In this paper, we propose a generalizable and efficient LLM-based recommendation framework RecCocktail. Our approach begins with fine-tuning a "base spirit" LoRA module using domain-general recommendation instruction data to align LLM with recommendation knowledge. Next, given users' behavior of a specific domain, we construct a domain-specific "ingredient" LoRA module. We then provide an entropy-guided adaptive merging method to mix the "base spirit" and the "ingredient" in the weight space. Please note that, RecCocktail combines the advantages of the existing two paradigms without introducing additional time or space overhead during the inference phase. Moreover, RecCocktail is efficient with plug and play, as the "base spirit" LoRA is trained only once, and any domain-specific "ingredient" can be efficiently mixed with only domain-specific fine-tuning. Extensive experiments on multiple datasets under both warm and cold-start recommendation scenarios validate the effectiveness and generality of the proposed RecCocktail.
Multimodal graphs, where nodes contain heterogeneous features such as images and text, are increasingly common in real-world applications. Effectively learning on such graphs requires both adaptive intra-modal message passing and efficient inter-modal aggregation. However, most existing approaches to multimodal graph learning are typically extended from conventional graph neural networks and rely on static structures or dense attention, which limit flexibility and expressive node embedding learning. In this paper, we propose a novel multimodal graph representation learning framework with Dynamic information Pathways (DiP). By introducing modality-specific pseudo nodes, DiP enables dynamic message routing within each modality via proximity-guided pseudo-node interactions and captures inter-modality dependence through efficient information pathways in a shared state space. This design achieves adaptive, expressive, and sparse message propagation across modalities with linear complexity. We conduct the link prediction and node classification tasks to evaluate performance and carry out full experimental analyses. Extensive experiments across multiple benchmarks demonstrate that DiP consistently outperforms baselines.