Aentive Collaborative Filtering: Multimedia Recommendation with Item- and Component-Level Aention

Multimedia content is dominating today’s Web information. Œe nature of multimedia user-item interactions is 1/0 binary implicit feedback (e.g., photo likes, video views, song downloads, etc.), which can be collected at a larger scale with a much lower cost than explicit feedback (e.g., product ratings). However, the majority of existing collaborative €ltering (CF) systems are not well-designed for multimedia recommendation, since they ignore the implicitness in users’ interactions with multimedia content. We argue that, in multimedia recommendation, there exists itemand component-level implicitness which blurs the underlying users’ preferences. Œe item-level implicitness means that users’ preferences on items (e.g., photos, videos, songs, etc.) are unknown, while the componentlevel implicitness means that inside each item users’ preferences on di‚erent components (e.g., regions in an image, frames of a video, etc.) are unknown. For example, a “view” on a video does not provide any speci€c information about how the user likes the video (i.e., item-level) and which parts of the video the user is interested in (i.e., component-level). In this paper, we introduce a novel aˆention mechanism in CF to address the challenging itemand component-level implicit feedback in multimedia recommendation, dubbed AŠentive Collaborative Filtering (ACF). Speci€cally, our aŠention model is a neural network that consists of two aŠention modules: the component-level aŠention module, starting from any content feature extraction network (e.g., CNN for images/videos), which learns to select informative components of multimedia items, and the item-level aŠention module, which learns to score the item preferences. ACF can be seamlessly incorporated into classic CF models with implicit feedback, such as BPR and SVD++, and eciently trained using SGD. Œrough extensive experiments on two real-world multimedia Web services: Vine and Pinterest, we show that ACF signi€cantly outperforms state-of-the-art CF methods. ∗Xiangnan He is the corresponding author. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for pro€t or commercial advantage and that copies bear this notice and the full citation on the €rst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permiŠed. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior speci€c permission and/or a fee. Request permissions from [email protected]. SIGIR’17, August 7–11, 2017, Shinjuku, Tokyo, Japan. © 2017 ACM. 978-1-4503-5022-8/17/08. . .$15.00 DOI: hŠp://dx.doi.org/10.1145/3077136.3080797 CCS CONCEPTS •Information systems →Multimedia information systems;