Recommender systems have long struggled with challenges such as cold start and data sparsity, which can lead to poor recommendation performance. While previous approaches have attempted to address these issues by incorporating side information, they often introduce noise, lack flexibility for data expansion, and suffer from inconsistent data quality—factors that hinder accurate user preference inference and reduce recommendation performance. With the vast knowledge bases and advanced reasoning capabilities of large language models (LLMs), these models are particularly well-suited to supplement auxiliary information and capture implicit user intent. To address these challenges, we propose a novel framework, ER2ALM, which leverages the capabilities of LLMs enhanced by Retrieval-Augmented Generation (RAG) to improve recommendation outcomes. Our framework specifically addresses the challenges by flexibly and accurately augmenting auxiliary information and capturing users’ implicit preferences and interests. Additionally, to mitigate the risk of introducing noise, we incorporate a noise reduction strategy to ensure the reliability of the augmented information. Experimental validation on two real-world datasets demonstrates the efficacy of our approach, significantly enhancing both the accuracy and robustness of recommendations compared to state-of-the-art methods. This demonstrates the potential of our framework as a new paradigm for preference mining in recommendation systems.

Online reviews on platforms such as Amazon and Yelp significantly influence consumer decisions and business reputations. However, spammers form groups that strategically control product reviews over specific periods to manipulate consumer sentiment and decision-making. Traditional spammer group detection methods face two primary issues: 1) knowledge marginalization: interactions dominate the model to form candidate groups, potentially marginalizing valuable structured knowledge; and 2) temporal knowledge discrepancy: inconsistencies in users’ temporal activities and behavioral features lead to blurry classification of candidate groups. To address these two limitations, we introducetemporal knowledge consistency for spammer groups detection via contrastive learningcalled TKCCL. We employ dual-view encoders derived from knowledge graphs and heterogeneous information networks to learn informative representations, thereby alleviating knowledge marginalization. TKCCL maps the temporal knowledge as vectors to measure consistency, enhancing users’ rating proximity and temporal synchronization in dual views, thereby reducing temporal knowledge discrepancies. We optimize spammer group detection by modeling it as a greedy set cover problem, which enhances the method’s responsiveness to dynamic spam strategies. Experimental results on four public datasets demonstrate that TKCCL substantially outperforms existing methods. Our code is available athttps://github.com/NeenLee/TKCCL.

Cross-domain recommendation aims to leverage knowledge from multiple domains to mitigate issues of data sparsity and cold–start problems. While traditional cross-domain settings often involve semantic domains (e.g., movies versus books), recent research has expanded this notion to include structure-level domains that reflect different types of interaction graphs. To this end, we proposeSuper–Item Interaction with Contrastive Learning for Structure-level Cross-domain Recommendation(SI${}^{2}$CL), a framework designed to explore item-level structural associations and transfer knowledge across graph-based domains. Specifically, SI${}^{2}$CL integrates a user–item interaction graph and an item–item graph—each capturing explicit and implicit signals, respectively—while addressing the challenge of noisy or sparse connections via contrastive denoising. Super–item interaction further facilitates knowledge transfer by modeling shared preferences of highly connected items and clusters. By reconstructing an item–item graph and aligning it with user feedback through structure-aware contrastive learning, our approach uncovers latent item relationships and improves recommendation robustness. Extensive experiments validate the effectiveness of SI${}^{2}$CL in enhancing both accuracy and diversity in sparse or cold–start recommendation scenarios.

Graph Neural Networks (GNNs) have demonstrated strong representation learning capabilities in recommender systems, particularly under the contrastive learning paradigm, where the construction of positive and negative sample pairs effectively captures latent relations between users and items, thereby significantly enhancing recommendation performance. However, existing graph contrastive learning methods predominantly rely on static augmentation strategies, lacking adaptability to diverse user behaviors and semantic structures. Moreover, effectively integrating external knowledge (e.g., user attributes and item semantics) into the contrastive learning process remains a major challenge. To address these limitations, we propose ProGraph, a graph prompt tuning framework tailored for recommendation tasks. ProGraph introduces adaptive contrastive learning within the graph prompt mechanism, enhanced by knowledge-aware guidance, to improve both the discriminability and semantic generalization of learned representations. Specifically, it employs structured prompts to guide GNNs in learning embeddings across multiple semantic subspaces, while incorporating knowledge-assisted graph views to preserve structural consistency and better handle heterogeneous attributes. Unlike traditional full-parameter optimization, ProGraph enables efficient tuning with a small number of learnable prompt parameters, thus achieving better transferability and modular compatibility. Extensive experiments on three real-world recommendation datasets with rich interaction records and knowledge attributes demonstrate that ProGraph consistently outperforms several representative state-of-the-art baselines in top-K recommendation performance.

Feature selection, an essential technique in data mining, is often confined to batch learning or online idealization of data scenarios despite its significance. Existing online feature selection methods have specific assumptions regarding the data stream, such as requiring a fixed feature space with an explicit pattern and complete labeling of samples. Unfortunately, data streams generated in many real scenarios commonly exhibit arbitrarily incomplete feature spaces and scarcity labels, making existing approaches unsuitable for real applications. To fill these gaps, this study proposes a new problem called Online Feature Selection with Varying Features Spaces (OFSVF). OFSVF has a three-fold main idea: 1) it leverages Gaussian Copula to model the incomplete feature correlation in a complete latent space, encoded by continuous variables, 2) it employs a novel tree-ensemble-based approach to select the most informative features on-the-fly, and 3) it develops the underlying geometric structure of instances to establish the relationship between unlabeled and labels. Experimental results are documented to demonstrate the feasibility and effectiveness of our proposed method.

Recommender systems have long struggled with challenges such as cold start and data sparsity, which can lead to poor recommendation performance. While previous approaches have attempted to address these issues by incorporating side information, they often introduce noise, lack flexibility for data expansion, and suffer from inconsistent data quality—factors that hinder accurate user preference inference and reduce recommendation performance. With the vast knowledge bases and advanced reasoning capabilities of large language models (LLMs), these models are particularly well-suited to supplement auxiliary information and capture implicit user intent. To address these challenges, we propose a novel framework, ER2ALM, which leverages the capabilities of LLMs enhanced by Retrieval-Augmented Generation (RAG) to improve recommendation outcomes. Our framework specifically addresses the challenges by flexibly and accurately augmenting auxiliary information and capturing users’ implicit preferences and interests. Additionally, to mitigate the risk of introducing noise, we incorporate a noise reduction strategy to ensure the reliability of the augmented information. Experimental validation on two real-world datasets demonstrates the efficacy of our approach, significantly enhancing both the accuracy and robustness of recommendations compared to state-of-the-art methods. This demonstrates the potential of our framework as a new paradigm for preference mining in recommendation systems.

With the growing privacy and data contamination concerns in recommendation systems, recommendation unlearning, i.e., unlearning the impact of specific learned data, has garnered more attention. Unfortunately, existing research primarily focuses on the complete unlearning of target data, neglecting the balance between unlearning integrity, practicality, and efficiency. Two major restrictions hinder the widespread application of this unlearning paradigm in practice. First, while prior studies often assume consistent similarity among samples, they overly emphasize the local collaborative relationships between samples and central nodes, leading to an imbalance between local and global collaborative information. Second, while data partition appears to be a default setup, this evidently exacerbates the sparsity of recommendation data, which can have a potentially negative impact on recommendation quality. To fill these gaps, this paper proposes a data partitioning and submodel training strategy, named Partition Distinction with Contrastive Recommendation Unlearning (PDCRU), which aims to balance data partitioning and feature sparsity. The key idea is to extract structural features as global collaborative information for samples and introduce structural feature constraints based on sample similarity during the partitioning process. For submodel training, we leverage contrastive learning to introduce additional high-quality training signals to enhance model embeddings. Extensive experiments validate the feasibility and consistent superiority of our method over existing recommendation unlearning models in learning and unlearning. Specifically, our model achieves a 4.83% improvement in performance and a 4.64x enhancement in unlearning efficiency compared to baseline methods. The code is released at https://github.com/linli0818/PDCRU.

The Uniform Resource Locator (URL) is a primary vector for numerous security threats, including phishing, malware propagation, and spam attacks, making URL-based analysis a critical task in security systems. However, existing research often focuses on static lexical features of individual URLs, overlooking deeper semantic, structural, and behavioral signals that can indicate malicious intent or evasive patterns. In this paper, we propose Behavior-Enhanced Semantic URL Embedding, a novel framework that integrates semantic, structural, and contextual information to improve the detection of security threats embedded in URLs. Our model is composed of three core modules: a semantic understanding module to extract token-level and contextual semantics, a topology structure learning module to capture hierarchical and sequential patterns of URL components, and a downstream multi-task adaptation module that fine-tunes embeddings with supervised contrastive learning for various security detection tasks. We evaluate our method across five public datasets covering key security applications such as malicious URL detection, phishing website identification, and spam filtering, consistently achieving superior performance over existing baselines. Additionally, we demonstrate the extensibility of our approach to related security tasks, showcasing its potential integration into real-world threat detection and security monitoring systems.

Online learning, characterized by its feature space’s adaptability over time, has emerged as a flexible learning paradigm that has attracted widespread attention. However, existing online learning methods often overlook the distributional differences between instances and the presence of label noise in streaming data, thus significantly hindering the effectiveness and robustness of these algorithms. To overcome these challenges, we propose an online confidence learning algorithm for noisy labeled features, which aims to achieve robustness against arbitrary data streams and noisy labels. It employs two new strategies: online confidence inference, which applies the principle of empirical risk minimization to identify inconsistencies in spatial distributions, and geometric structure learning, which utilizes dynamic instance confidence to compute disparities between instances and their labels. Empirical findings demonstrate that our label correction mechanism enhances classification accuracy more effectively across various types of noisy labels (i.e., symmetric, asymmetric, and flipped). Additionally, a case study on image datasets was conducted to illustrate in detail the effectiveness of our OLNLS algorithm. Code is released in https://github.com/Zhuosd/OLNLS .

Online learning, where feature spaces can change over time, offers a flexible learning paradigm that has attracted considerable attention. However, it still faces three significant challenges. First, the heterogeneity of real-world data streams with mixed feature types presents challenges for traditional parametric modeling. Second, data stream distributions can shift over time, causing an abrupt and substantial decline in model performance. Additionally, the time and cost constraints make it infeasible to label every data instance in a supervised setting. To overcome these challenges, we propose a new algorithm Online Learning from Mix-typed, Drifted, and Incomplete Streaming Features (OL-MDISF), which aims to relax restrictions on both feature types, data distribution, and supervision information. Our approach involves utilizing copula models to create a comprehensive latent space, employing an adaptive sliding window for detecting drift points to ensure model stability, and establishing label proximity information based on geometric structural relationships. To demonstrate the model’s efficiency and effectiveness, we provide theoretical analysis and comprehensive experimental results.

The advancement of intelligent systems, particularly in domains such as natural language processing and autonomous driving, has been primarily driven by deep neural networks (DNNs). However, these systems exhibit vulnerability to adversarial attacks that can be both subtle and imperceptible to humans, resulting in arbitrary and erroneous decisions. This susceptibility arises from the hierarchical layer‐by‐layer learning structure of DNNs, where small distortions can be exponentially amplified. While several defense methods have been proposed, they often necessitate prior knowledge of adversarial attacks to design specific defense strategies. This requirement is often unfeasible in real‐world attack scenarios. In this paper, we introduce a novel learning model, termed “immune” learning, known as adversarial‐resilient deep symbolic tree (ARDST), from a neurosymbolic perspective. The ARDST model is semiparametric and takes the form of a tree, with logic operators serving as nodes and learned parameters as weights of edges. This model provides a transparent reasoning path for decision‐making, offering fine granularity, and has the capacity to withstand various types of adversarial attacks, all while maintaining a significantly smaller parameter space compared to DNNs. Our extensive experiments, conducted on three benchmark datasets, reveal that ARDST exhibits a representation learning capability similar to DNNs in perceptual tasks and demonstrates resilience against state‐of‐the‐art adversarial attacks.

The blockchain can offer a dependable and secure platform for Internet of Things (IoT) transactions with its distributed and secure network architecture. Unfortunately, it faces challenges, such as limited throughput, excessive computational costs, and high-transaction fees. Off-chain scaling protocols are used to address the scalability of blockchain for their outstanding performance and efficiency. To mitigate the high-cost interactions with blockchain, previous studies only considered moving transactions of payment hubs (PHs) off-chain, utilizing off-chain operators to aggregate multiple transactions. However, existing PHs overly rely on central operators for system maintenance, greatly increasing the risk of central operator failure (COF). Previous solutions allowed operators to submit unsettled state commitments (USCs) to the blockchain and overlooked the pessimistic scenario that could lead to state rollbacks. To address these issues, this article proposes an efficient multiparty payment protocol (HyperPay), aimed at utilizing the off-chain scaling technique to enhance transaction throughput and reduce on-chain cost. Specifically, we first propose a novel off-chain committee and collateral-based verifiable random leader election (C-VRE) to elect leaders fairly, thus mitigating the COF problem. Additionally, we design a new state validation mechanism and one-step fraud challenge (OSFC), enabling verifiers to directly construct fraud proofs and challenges on-chain, thereby preventing leaders from submitting USC. Our evaluation indicates that HyperPay reduces on-chain costs of challenge by 80% and boosts peak throughput by a factor of 10X-283X. A comprehensive theoretical analysis and experimental results substantiate the security and effectiveness of our proposed approach.

Feature selection, an essential technique in data mining, is often confined to batch learning or online idealization of data scenarios despite its significance. Existing online feature selection methods have specific assumptions regarding the data stream, such as requiring a fixed feature space with an explicit pattern and complete labeling of samples. Unfortunately, data streams generated in many real scenarios commonly exhibit arbitrarily incomplete feature spaces and scarcity labels, making existing approaches unsuitable for real applications. To fill these gaps, this study proposes a new problem calledOnline Feature Selection with Varying Features Spaces(OFSVF). OFSVF has a three-fold main idea: 1) it leverages Gaussian Copula to model the incomplete feature correlation in a complete latent space, encoded by continuous variables, 2) it employs a novel tree-ensemble-based approach to select the most informative features on-the-fly, and 3) it develops the underlying geometric structure of instances to establish the relationship between unlabeled and labels. Experimental results are documented to demonstrate the feasibility and effectiveness of our proposed method.

Online learning with feature spaces that are not fixed but can vary over time renders a seemingly flexible learning paradigm thus has drawn much attention. Unfortunately, two restrictions prohibit a ubiquitous application of this learning paradigm in practice. First, whereas prior studies mainly assume a homogenous feature type, data streams generated from real applications can be heterogeneous in which Boolean, ordinal, and continuous co-exist. Existing methods that prescribe parametric distributions such as Gaussians would not suffice to model the correlation among such mixtyped features. Second, while full supervision seems to be a default setup, providing labels to all arriving data instances over a long time span is tangibly onerous, laborious, and economically unsustainable. Alas, a semi-supervised online learner that can deal with mix-typed, varying feature spaces is still missing. To fill the gap, this paper explores a novel problem, named Online Semi-supervised Learning with Mixtyped streaming Features (OSLMF), which strives to relax the restrictions on the feature type and supervision information. Our key idea to solve the new problem is to leverage copula model to align the data instances with different feature spaces so as to make their distance measurable. A geometric structure underlying data instances is then established in an online fashion based on their distances, through which the limited labeling information is propagated, from the scarce labeled instances to their close neighbors. Experimental results are documented to evidence the viability and effectiveness of our proposed approach. Code is released in https://github.com/wudi1989/OSLMF.

Knowledge graph (KG) as auxiliary information can solve the cold-start and data sparsity problems of recommender systems.However, most existing KG-based recommendation methods focus on how to effectively encode items with that users have interacted into entities and propagate them explicitly, but neglect the relation-level and context-level modeling of collaborative signals.Therefore, it is inevitable to incorporate some unrelated entities while utilizing a propagation strategy, which may weaken part of the recommendation performance.To address this problem, we propose a novel method named Collaborative Relation Context Consistency (CRCC).Compared with other KG-based methods, we model the relation-level and context-level of collaborative signals in a fine-grained manner.Specifically, we segment the user's collaborative knowledge graph to learn related entity information separately to enrich the embedding of users.Moreover, CRCC links the consistency score between the items that users and neighbors have interacted with as the fusion basis, and then we consider the inherent popularity of items while incorporating consistent entities to enhance the embedding representation of items.Extensive experiments on three real-world datasets show that CRCC outperforms several compelling baselines in both CTR prediction and top-K recommendation.