超越单一视角：针对推荐系统中稀疏且不完整的用户生成内容的元评审
Beyond a Single Story: Meta-Reviewing Sparse and Incomplete User-generated Contents for Recommendation
浏览论文内容
中文总结 AI 辅助
本文针对推荐系统中用户生成内容的稀疏与不完整问题，提出MOSAIC模型，通过聚合邻居用户评论的属性-情感证据构建元评审，结合MMoE与注意力模块提升推荐准确率与解释质量，在多数据集上优于现有基线。
中文摘要 AI 辅助
数据稀疏性是推荐系统长期存在的挑战，对于依赖用户生成内容（UGC，如文本评论）的方法而言，该问题更为严峻——UGC能捕捉细粒度偏好，但需要用户付出更多努力生成。因此，UGC存在两类问题：（1）缺失评论，即交互行为无对应评论；（2）不完整评论，即现有评论仅覆盖相关属性的子集。现有方法常忽视UGC特有的这些问题，导致准确率下降。受学术同行评审中“元评审”的启发，本文提出MOSAIC（Meta-review On Sparse And Incomplete user-generated Content），通过聚合邻居用户评论中的属性-情感证据，为每个目标用户构建元评审。多门专家混合（MMoE）架构联合优化评分预测与元评审属性-情感预测，同时注意力模块将聚合的元评审信号个性化适配至每个目标用户，从而得到更精准的评分预测及属性级解释。在四个真实数据集上的实验表明，MOSAIC在推荐准确率和解释质量上均持续优于当前最优基线方法，缓解了UGC的稀疏性与不完整性问题，且为交互历史有限的用户带来了持续的性能提升。
英文摘要
Data sparsity remains a long-standing challenge in recommender systems, and it becomes more severe for methods relying on user-generated content (UGC) such as textual reviews, which capture fine-grained preferences but require more user efforts to produce. As a result, UGC exhibits (1) missing reviews, where interactions lack any review, and (2) incomplete reviews, where available reviews cover only a subset of relevant attributes. Existing approaches often overlook these UGC-specific issues, leading to degraded accuracy. Motivated by meta-review in academic peer review, we propose MOSAIC (Meta-review On Sparse And Incomplete user-generated Content), which constructs a meta-review for each target user by aggregating attribute-sentiment evidence from neighbor users' reviews. A multi-gate mixture-of-experts (MMoE) architecture jointly optimizes rating prediction and meta-review attribute-sentiment prediction, while an attention module personalizes the aggregated meta-review signals to each target user, yielding both refined rating predictions and attribute-level explanations. Experiments on four real-world datasets demonstrate that MOSAIC consistently outperforms state-of-the-art baselines in both recommendation accuracy and explanation quality, mitigating UGC sparsity and incompleteness while delivering consistent gains for users with limited interaction history.
