Overhead-free User-side Recommender Systems

TL;DR

RecCycle is proposed, which realizes user-side recommender systems without any communication overhead and performs as well as state-of-the-art user-side recommendation algorithms while RecCycle reduces costs significantly.

Abstract

Traditionally, recommendation algorithms have been designed for service developers. But recently, a new paradigm called user-side recommender systems has been proposed. User-side recommender systems are built and used by end users, in sharp contrast to traditional provider-side recommender systems. Even if the official recommender system offered by the provider is not fair, end users can create and enjoy their own user-side recommender systems by themselves. Although the concept of user-side recommender systems is attractive, the problem is they require tremendous communication costs between the user and the official system. Even the most efficient user-side recommender systems require about 5 times more costs than provider-side recommender systems. Such high costs hinder the adoption of user-side recommender systems. In this paper, we propose overhead-free user-side recommender systems, RecCycle, which realizes user-side recommender systems without any communication overhead. The main idea of RecCycle is to recycle past recommendation results offered by the provider's recommender systems. The ingredients of RecCycle can be retrieved ``for free,'' and it greatly reduces the cost of user-side recommendations. In the experiments, we confirm that RecCycle performs as well as state-of-the-art user-side recommendation algorithms while RecCycle reduces costs significantly.

Figures (3)

• Figure 1: Illustration of the cached recommendation network. When the user visits a page represented by a node, surrounding nodes are recommended and observed in the DOM. The boundary items of the visited items can be reached by the search in the cached recommendation network. These items are likely to be relevant to the user. There are clusters that the user has never visited, and the cached graph does not contain them, but this is not a problem for effective recommendations in most cases.
• Figure 2: Table 3. Performance Comparison. Cost denotes the average number of times each method accesses item pages, i.e., the number of queries to the official recommender systems. The less this value is, the more communication-efficient the method is. The best score is highlighted in bold among the four methods (excluding the Oracle method). RecCycle is extremely more efficient than other methods by achieving zero communication overhead (i.e., it accesses only one page, the item page the user is currently viewing) while it achieves on par or slightly worse performances than the Oracle method and state-of-the-art user-side recommender systems.
• Figure 2: Performance of RecCycle with different lengths of user histories. Note that the y-axis starts at around $0.102$ for recall and $0.058$ for nDCG, and the relative difference in performance is within a few percent. RecCycle achieves good performance even when the user history is short and the cache is sparse.

Theorems & Definitions (1)

• Theorem 1