recommendation algorithms

Existing online sequential recommendation methods underutilize structural information from both users and items, typically relying on only one side (e.g., user- or item-only structures), leading to suboptimal performance. Method: We propose the first sequential recommendation framework that jointly models *both* user-side and item-side type structures. Leveraging latent-variable clustering, it constructs dual-sided type structures and operates under an i.i.d. type-preference assumption. Contribution/Results: Through information-theoretic analysis, we prove the inherent suboptimality of single-sided structural modeling and establish the first near information-theoretically optimal sequential recommendation algorithm. Our method achieves cumulative regret asymptotically approaching the theoretical lower bound—significantly outperforming classical single-structure baselines across standard benchmarks.

This work addresses the challenge of effectively integrating user–item and item–item collaborative filtering to enhance Top-N recommendation performance while maintaining computational efficiency. The authors propose a weighted similarity ensemble method based on shared embeddings, which, for the first time, unifies both recommendation pathways within a single framework. By sharing user and item embeddings across strategies, the approach simplifies model architecture and eliminates the need for separate hyperparameter tuning for each pathway, thereby substantially reducing deployment complexity. Experimental results demonstrate that the proposed method achieves competitive recommendation accuracy across multiple datasets and exhibits robust performance in scenarios favoring different collaborative filtering paradigms.

This paper addresses core challenges in leveraging user reviews for recommendation systems—namely, inadequate review text modeling, weak integration with explicit ratings, and insufficient interpretability—and proposes the first unified taxonomy for review-enhanced recommendation, systematically surveying representative works from 2014 to 2024. Methodologically, it integrates BERT/LSTM encoders, graph neural networks, attention mechanisms, and multi-task learning to jointly model review semantics, user-item interactions, and fine-grained features (e.g., attribute-level preferences). Its contributions are threefold: (1) it formally defines three emerging research frontiers—multimodal fusion, multi-criteria rating modeling, and ethics-aligned recommendation; (2) it identifies critical bottlenecks in model generalizability, robustness to sparse reviews, and attribution-based interpretability; and (3) it outlines theoretically grounded yet practically feasible future research directions.

To address the “paradox of choice” induced by content overload in video-on-demand platforms, this paper proposes an implicit-feedback-based graph neural recommendation method. We construct a user–video bipartite interaction graph where edge weights reflect behavioral signals—such as viewing percentage—and integrate degree, closeness, and betweenness centrality to quantify node importance. Crucially, we introduce the first integration of Louvain modularity clustering with a user-centric graph ranking mechanism, enabling personalized recommendations without reliance on explicit ratings. This approach overcomes the fundamental limitation of conventional collaborative filtering methods that depend on explicit feedback, thereby substantially improving recommendation accuracy. Empirical evaluation on a documentary streaming platform demonstrates significant performance gains: click-through rate increases by 63%, completion rate by 24%, and user satisfaction by 17%, all outperforming baseline models—including Naïve Bayes and SVM—by substantial margins.

Cold-start recommendation remains a fundamental challenge, as conventional approaches rely on initial user interactions or item attributes—while popularity- or random-based baselines suffer from poor accuracy. This paper proposes the first zero-data initialization framework for recommender systems: it requires no user behavior logs or item metadata, instead leveraging implicit structural priors and zero-initialized tensor optimization to perform unsupervised matrix completion, augmented with fairness-aware regularization. Under standard evaluation protocols, our method achieves MAE comparable to fully supervised matrix factorization baselines trained on abundant data, substantially outperforming random recommendations. Moreover, fairness metrics—measured via demographic parity and equalized odds—show marked improvement over existing cold-start methods. Crucially, the framework eliminates dependence on cold-start data collection entirely, establishing a novel paradigm for cold-start recommendation grounded in structural self-supervision and fairness-constrained optimization.

Existing recommender system research lacks a rigorous definition, structural characterization, and mechanistic analysis of collaborative information. Method: This paper introduces the first quantitative definition of collaborative information based on item co-occurrence patterns, systematically uncovering its distributional regularities and structural properties in user–item interaction data. Through co-occurrence pattern analysis, multi-dimensional statistical modeling, and cross-algorithm empirical evaluation—including collaborative filtering and graph neural networks—we rigorously assess its impact on recommendation performance. Contribution/Results: We demonstrate that collaborative information density and heterogeneity significantly and differentially drive model generalization capability and long-tail item coverage. Our work establishes the first empirically grounded analytical framework for collaborative information, offering interpretable optimization pathways and theoretical foundations for recommender algorithm design.

Academic recommendation research has long been constrained by offline, small-scale datasets and idealized evaluation protocols, failing to reflect real-world industrial constraints. To address this gap, we conduct a systematic literature review, industrial case studies, and cross-domain comparative analysis—thereby proposing, for the first time, a dichotomous taxonomy grounded in item attributes and recommendation objectives: *transactional* (emphasizing conversion efficiency and economic rationality) versus *content-oriented* (focusing on interest modeling and cognitive mechanisms). Integrating theories from user decision psychology and microeconomics, our framework clarifies core challenges—including real-time responsiveness, data sparsity, and dynamic feedback—and exposes structural disparities between academia and industry in data scale, update frequency, and evaluation paradigms. This work establishes a conceptual foundation and technical roadmap for bridging the research-practice divide, advancing recommendation systems toward interpretability, deployability, and empirical verifiability.

Offline evaluation in recommender systems often fails to accurately reflect users’ true preferences due to data sparsity and various biases. This study systematically investigates how different offline evaluation designs—such as data filtering thresholds and candidate set construction strategies—affect model ranking outcomes. For the first time, it quantifies the convergent validity of these designs by measuring their ranking consistency against rankings derived from dense, real user feedback. Through extensive multi-configuration experiments and correlation analyses, the work demonstrates that the effectiveness of offline evaluation is highly dependent on the specific dataset and target task, with no universally optimal design. These findings underscore a critical principle: evaluation protocols must be carefully aligned with the intended application context.

Existing offline evaluation metrics for fairness in recommender systems lack systematic analysis of robustness, interpretability, and applicability, making it difficult to select and interpret them appropriately in practice. This work presents the first comprehensive examination of mainstream fairness metrics, uncovering their theoretical shortcomings and empirical limitations across user- and item-side perspectives as well as group- and individual-level granularities. By integrating multidimensional fairness definitions and employing theoretical derivations, distributional analyses, and boundary-case testing, the study delineates the operational boundaries and failure modes of these metrics. Building on these insights, the paper proposes refined evaluation methodologies and practical guidelines for metric selection, substantially enhancing the reliability and actionable utility of fairness assessments in recommender systems.

This paper investigates whether weighted matrix factorization (WMF) improves recommendation performance in implicit-feedback settings. Through systematic analysis of the coupling effects among weighting schemes, model capacity, and regularization, we find that unweighted training achieves performance comparable to—or even surpassing—that of state-of-the-art weighted methods under high-capacity models, challenging the conventional assumption that weighting universally enhances performance. To address the computational difficulty of exact optimization for classical weighted objectives (e.g., WALS), we propose a novel, efficient, and provably exact optimization algorithm—the first to minimize such non-convex, non-smooth weighted objectives without approximation. Extensive experiments across diverse MF architectures and weighting strategies on multiple benchmark datasets confirm our findings: weighting yields gains only under low-capacity models or strong regularization, whereas simplified unweighted training is more robust and efficient for large-scale models. Our core contributions are (i) identifying precise boundary conditions under which weighting is beneficial, (ii) providing theoretical justification, and (iii) delivering a practical, exact optimization tool.