Large Language Models (LLMs) have been widely adopted in ranking systems such as information retrieval (IR) systems and recommender systems (RSs). To alleviate the latency of auto-regressive decoding, some studies explore the single (first) token decoding for ranking approximation, but they suffer from severe degradation in tail positions. Although speculative decoding (SD) methods can be a remedy with verification at different positions, they face challenges in ranking systems due to their left-to-right decoding paradigm. Firstly, ranking systems require strict latency constraints, but verification rounds in SD methods remain agnostic; Secondly, SD methods usually discard listwise ranking knowledge about unaccepted items in previous rounds, hindering future multi-token prediction, especially when candidate tokens are the unaccepted items. In this paper, we propose a Reinforcement Speculative Decoding method for fast ranking inference of LLMs. To meet the ranking systems' latency requirement, we propose an up-to-down decoding paradigm that employs an agent to iteratively modify the ranking sequence under a constrained budget. Specifically, we design a ranking-tailored policy optimization, actively exploring optimal multi-round ranking modification policy verified by LLMs via reinforcement learning (RL). To better approximate the target LLM under the constrained budget, we trigger the agent fully utilizing the listwise ranking knowledge about all items verified by LLMs across different rounds in RL, enhancing the modification policy of the agent. More importantly, we demonstrate the theoretical robustness and advantages of our paradigm and implementation. Experiments on both IR and RS tasks show the effectiveness of our proposed method.

Large language models (LLMs) are increasingly applied to ranking tasks in retrieval and recommendation. Although reasoning prompting can enhance ranking utility, our preliminary exploration reveals that its benefits are inconsistent and come at a substantial computational cost, suggesting that when to reason is as crucial as how to reason. To address this issue, we propose a reasoning routing framework that employs a lightweight, plug-and-play router head to decide whether to use direct inference (Non-Think) or reasoning (Think) for each instance before generation. The router head relies solely on pre-generation signals: i) compact ranking-aware features (e.g., candidate dispersion) and ii) model-aware difficulty signals derived from a diagnostic checklist reflecting the model's estimated need for reasoning. By leveraging these features before generation, the router outputs a controllable token that determines whether to apply the Think mode. Furthermore, the router can adaptively select its operating policy along the validation Pareto frontier at deployment time, enabling dynamic allocation of computational resources toward instances most likely to benefit from Think under varying system constraints. Experiments on three public ranking datasets with different scales of open-source LLMs show consistent improvements in ranking utility with reduced token consumption (e.g., +6.3% NDCG@10 with –49.5% tokens on MovieLens with Qwen3-4B), demonstrating reasoning routing as a practical solution to the accuracy-efficiency trade-off.

Large Language Models (LLMs) have been widely adopted in ranking systems, specifically for reranking tasks. Despite the effectiveness, the auto-regressive decoding of LLMs leads to inference latency due to the memory-bandwidth-bound. To alleviate this bottleneck, prior studies have explored single token decoding as an approximation, but they suffer from performance degradation at the tail positions of ranking. In this paper, we propose a Single Forward Pass (SFP)-based method to pre-verify multiple rankings using tree attention, approximating auto-regressive decoding by relevant sub-rankings at each step. However, verifying all possible ranking permutations will lead to factorial-level token computation (N!), making it intractable within SFP. To this end, we first reduce item ranking permutations to combinations (2N), based on the empirical observation that the LLM's next-item generation is less sensitive to the exact ordering of preceding items. Furthermore, we divide the full ranking into K sub-rankings and aggregate their individual probabilities, which further decreases the verification space to 2⌈N/K⌉ • K << 2N. However, naïvely aggregating sub-ranking probabilities leads to inaccurate estimates in listwise ranking. To overcome this, we introduce a Möbius inversion model that explicitly decompose the individual contribution of subsets within a complete lattice, as verified by tree attention. Then, we learn their higher-order effects with a hierarchical self-attention model to reconstruct the full ranking probability. Experiments on both information retrieval and recommendation tasks show the effectiveness of our proposed method.

Expert-sharing patterns have emerged as promising paradigms for multi-task learning (MTL) in recommender systems, enabling efficient resource allocation and dynamic modeling of diverse tasks. In this paper, we observe that high-gating (leader) and low-gating (auxiliary) experts play distinct roles in MTL: leader experts dominate task-specific predictions, while auxiliary experts, despite their lower gating scores, often contain complementary knowledge that can enhance model performance. However, critical challenges persist: how to effectively identify and utilize the knowledge of the leader and auxiliary experts in a synergistic manner? To address this, we propose a novel Dynamic Experts Synergy (DES) mechanism that integrates Entropy-driven Experts Classification (EEC) and Multi-view Knowledge Recycle (MVKR). EEC dynamically partitions experts into leader and auxiliary groups by analyzing task-specific prediction and gating entropy, enabling adaptive allocation aligned with real-time task difficulty. MVKR effectively revisits knowledge from auxiliary experts through utility, diversity, and task-relatedness perspectives, ensuring comprehensive knowledge utilization. Extensive experiments on five datasets demonstrate the superiority of our DES against state-of-the-art methods.

Social recommendations (SRs) aim to enhance preference modeling by integrating social networks. However, their effectiveness is mainly constrained by two factors: the noisy social connections that may not reflect shared interests, and the limited number of social connections for most users, which hampers the system's ability to fully leverage social influence. Therefore, it is essential to perform social network refinement by removing noisy connections and adding meaningful ones for robust SRs. Inspired by the denoising capability of generative diffusion models, we propose a Model-Agnostic Social Network Refinement framework with Diffusion Models for Robust Social Recommendation (ARD-SR). Specifically, in the forward process, we corrupt the social network by progressively adding position-specific Gaussian noise calibrated to the user preference similarity, better simulating how the social network responds to noise perturbations. The reverse process learns to denoise, guided by each user's neighborhood preferences from the SR backbone, generating a tailored social network aligned with each user's preference for establishing connections. For effective learning, we design a curriculum-based training mechanism that progressively introduces challenging samples characterized by high sparsity or high noise levels. Finally, ARD-SR and the SR backbone are alternately trained, ensuring a continuous mutual enhancement between the social network refinement and the backbone's user representation learning. To further enhance the quality of the refined social network, (1) we introduce a preference-guided flip operation during inference to improve the input quality; and (2) we modify social connections based on the exponential weighted moving average of ARD-SR's predictions across epochs to reduce fluctuations. Experiments on three datasets show that ARD-SR significantly improves SR performance across multiple SR backbones. The code is released at https://github.com/sunyc123r/ARD-SR.

Recently, much effort has been devoted to modeling users' multi-interests (aka multi-faceted preferences) based on their behaviors, aiming to accurately capture users' complex preferences. Existing methods attempt to model each interest of users through a distinct representation, but these multi-interest representations easily collapse into similar ones due to a lack of effective guidance. In this paper, we propose a generic multi-interest method for sequential recommendation, achieving disentangled representation learning of diverse interests technically and theoretically. To alleviate the collapse issue of multi-interests, we propose to conduct item partition guided by their likelihood of being co-purchased in a global view. It can encourage items in each group to focus on a discriminated interest, thus achieving effective disentangled learning of multi-interests. Specifically, we first prove the theoretical connection between item partition and spectral clustering, demonstrating its effectiveness in alleviating item-level and facet-level collapse issues that hinder existing disentangled methods. To efficiently optimize this problem, we then propose a Markov Random Field (MRF)-based method that samples small-scale sub-graphs from two separate MRFs, thus it can be approximated with a cross-entropy loss and optimized through contrastive learning. Finally, we perform multi-task learning to seamlessly align item partition learning with multi-interest modeling for more accurate recommendation. Experiments on three real-world datasets show that our method significantly outperforms state-of-the-art methods and can flexibly integrate with existing multi-interest models as a plugin to enhance their performances.

Social Recommendation (SR) typically exploits neighborhood influence in the social network to enhance user preference modeling. However, users' intricate social behaviors may introduce noisy social connections for user modeling and harm the models' robustness. Existing solutions to alleviate social noise either filter out the noisy connections or generate new potential social connections. Due to the absence of labels, the former approaches may retain uncertain connections for user preference modeling while the latter methods may introduce additional social noise. Through data analysis, we discover that (1) social noise likely comes from the connected users with low preference similarity; and (2) Opinion Leaders (OLs) play a pivotal role in influence dissemination, surpassing high-similarity neighbors, regardless of their preference similarity with trusting peers. Guided by these observations, we propose a novel Self-Supervised Denoising approach through Independent Cascade Graph Augmentation, for more robust SR. Specifically, we employ the independent cascade diffusion model to generate an augmented graph view, which traverses the social graph and activates the edges in sequence to simulate the cascading influence spread. To steer the augmentation towards a denoised social graph, we (1) introduce a hierarchical contrastive loss to prioritize the activation of OLs first, followed by high-similarity neighbors, while weakening the low-similarity neighbors; and (2) integrate an information bottleneck based contrastive loss, aiming to minimize mutual information between original and augmented graphs yet preserve sufficient information for improved SR. Experiments conducted on two public datasets demonstrate that our model outperforms the state-of-the-art while also exhibiting higher robustness to different extents of social noise.

Most existing denoising recommendation methods alleviate noisy implicit feedback (user behaviors) through mainly empirical studies. However, such studies may lack theoretical explainability and fail to model comprehensive noise patterns, which hinders the understanding and capturing of different noise patterns that affect users’ behaviors. Thus, we propose to capture comprehensive noise patterns through theoretical and empirical analysis for more effective denoising, where users’ behaviors are divided into willingness and action phases to disentangle independent noise patterns. Willingness refers to the user’s intent to interact with an item, which may not lead to actual interaction due to different factors such as misclicking. Action denotes the user’s actual interaction with an item. Our analysis unveils that (1) in the willingness phase, high uncertainty in the user’s willingness to interact with the item can lead to high expectation loss which aligns with the findings of existing denoising methods; and (2) in the action phase, higher user-specific inconsistency between willingness and action not only leads to more noise in the user’s overall behaviors but also makes it harder to distinguish between true and noisy behaviors. Inspired by these findings, we propose a Unified Denoising Training (UDT) method for recommendation. To alleviate uncertainty in the willingness phase, we lower the importance of the user-item interaction with high willingness uncertainty recognized by high loss. To ease the inconsistency in the action phase, we lower the importance for users with high user-specific inconsistency as it may lead to noisier behaviors. Then, we increase the importance gap between the clean and noisy behaviors for users with low user-specific inconsistency as their behaviors are more distinguishable. Extensive experiments on three real-world datasets show that our proposed UDT outperforms state-of-the-art denoising recommendation methods.

Reciprocal recommendation is the core of many social websites like online recruitment and online dating. Most recently, graph neural networks have been exploited by few researchers for reciprocal recommendation. However, they tend to oversimplify the interactions between users, treating them as simple pairwise relationships, which overlooks the multidimensional relationships among users. Additionally, these methods fail to consider users’ historical interaction sequences and feedback behaviors, which makes it challenging to effectively capture the changes of user preferences over time.To address these issues, this study proposes a novel bilateral recommendation model based on sequential hypergraphs for reciprocal scenarios. Firstly, to capture the complex multidimensional relationships between bilateral users, we design a new data structure called bilateral sequential hypergraphs to capture the diverse relationships among users and to mine collaborative signals at the sequential level. Secondly, we propose corresponding bilateral sequential hypergraph convolution structures to learn the embedded representations of bilateral users. To adequately capture the changes in user preferences, the model incorporates position modeling and feedback behavior modeling within the proposed convolution strategy. Extensive experiments on several real-world datasets demonstrate that the proposed method consistently outperforms state-of-the-art methods.

Online reviews, which contain the quality information and user experience about products, always affect the consumption decisions of customers. Unfortunately, quite a number of spammers attempt to mislead consumers by writing fake reviews for some intents. Existing methods for detecting spam reviews mainly focus on constructing discriminative features, which heavily depend on experts and may miss some complex but effective features. Recently, some models attempt to learn the latent representations of reviews, users, and items. However, the learned embeddings usually lack interpretability. Moreover, most of existing methods are based on single classification model while ignoring the complementarity of different classification models.

Precise prediction of users' behavior is critical for users' satisfaction and platforms' benefit. A user's behavior heavily depends on the user's general preference and contextual information (current location, weather etc.). In this paper, we propose a succinct hierarchical framework named Hierarchical Hybrid Feature Model (HHFM). It combines users' general taste and diverse contextual information into a hybrid feature representation to profile users' dynamic preference w.r.t context. Meanwhile, we propose an n-way concatenation pooling strategy to capture the non-linear and complex inherent structures of real-world data, which were ignored by most existing methods like Factorization Machines. Conceptually, our model subsumes several existing methods when choosing proper concatenation and pooling strategies. Extensive experiments show our model consistently outperforms state-of-the-art methods on three real-world data sets.