用动态曲率的几何模型实现可解释的推荐推理
ManifoldMind: Dynamic Hyperbolic Reasoning for Trustworthy Recommendations
- 将用户、物品和标签建模为可变曲率的概率球体
- 在四个数据集上表现优于基线，提升准确率与多样性
- 生成清晰推理路径，适合需要可信推荐的场景
我们提出ManifoldMind，一种基于概率几何的推荐系统，用于在双曲空间中对语义层级进行探索性推理。与以往固定曲率和刚性嵌入的方法不同，ManifoldMind将用户、物品和标签表示为自适应曲率的概率球体，支持个性化不确定性建模和几何感知的语义探索。通过曲率感知的语义核，实现软性多跳推理，避免过度依赖浅层交互，探索更丰富的概念路径。在四个公开基准上的实验表明，其在NDCG、校准性和多样性方面均优于强基线。ManifoldMind能生成明确的推理轨迹，实现在稀疏或抽象领域中的透明、可信、探索驱动推荐。
原文摘要 · Abstract (English)
We introduce ManifoldMind, a probabilistic geometric recommender system for exploratory reasoning over semantic hierarchies in hyperbolic space. Unlike prior methods with fixed curvature and rigid embeddings, ManifoldMind represents users, items, and tags as adaptive-curvature probabilistic spheres, enabling personalised uncertainty modeling and geometry-aware semantic exploration. A curvature-aware semantic kernel supports soft, multi-hop inference, allowing the model to explore diverse conceptual paths instead of overfitting to shallow or direct interactions. Experiments on four public benchmarks show superior NDCG, calibration, and diversity compared to strong baselines. ManifoldMind produces explicit reasoning traces, enabling transparent, trustworthy, and exploration-driven recommendations in sparse or abstract domains.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。