Abstract
A series of graph filtering (GF)-based collaborative filtering (CF) showcases state-of-the-art performance on the recommendation accuracy by using a low-pass filter (LPF) without a training process. However, conventional GF-based CF approaches mostly perform matrix decomposition on the item-item similarity graph to realize the ideal LPF, which results in a non-trivial computational cost and thus makes them less practical in scenarios where rapid recommendations are essential. In this paper, we propose Turbo-CF, a GF-based CF method that is both training-free and matrix decomposition-free. Turbo-CF employs a polynomial graph filter to circumvent the issue of expensive matrix decompositions, enabling us to make full use of modern computer hardware components (i.e., GPU). Specifically, Turbo-CF first constructs an item-item similarity graph whose edge weights are effectively regulated. Then, our own polynomial LPFs are designed to retain only low-frequency signals without explicit matrix decompositions. We demonstrate that Turbo-CF is extremely fast yet accurate, achieving a runtime of less than 1 second on real-world benchmark datasets while achieving recommendation accuracies comparable to best competitors.
Abstract (translated)
一系列基于图滤波（GF）的协同过滤（CF）方法展示了一种无需训练过程在推荐准确性上实现最先进性能的方法。然而，传统的GF-based CF方法通常在物品物品相似图上进行矩阵分解，以实现理想的LPF，这导致计算成本非零，因此在需要快速推荐的场景中，它们变得更不实用。在本文中，我们提出了Turbo-CF，一种基于GF的CF方法，既无需训练过程又无需矩阵分解。Turbo-CF采用多项式图滤波器来绕过昂贵的矩阵分解问题，使我们能够充分利用现代计算机硬件组件（即GPU）。具体来说，Turbo-CF首先构建了一个物品物品相似图，其中边权重有效地调节。然后，我们自己的多项式低频信号保留函数被设计为仅保留低频信号，而没有显式矩阵分解。我们证明了Turbo-CF既快速又准确，在现实世界的基准数据集上的运行时间不到1秒，同时具有与最佳竞争对手相当的推荐准确性。
URL
https://arxiv.org/abs/2404.14243
