Learning aspect-level complementarity for intent-aware complementary recommendation

doi:10.1016/j.knosys.2022.109936

Return

Learning aspect-level complementarity for intent-aware complementary recommendation

2022-12-01

PRE

王锐 (Rui Wang)

Ning Yang *

Philip S. Yu

DOI：10.1016/j.knosys.2022.109936

Original

Original request for help

Save

Abstract

En 中文

Complementary recommendation aims to recommend items that are dissimilar but relevant to, and likely purchased together with, the items a user has purchased. Although many efforts have been made, the existing works for complementary recommendation still suffer from two shortcomings. First, the existing works often model the complementarity between items in terms of their co-occurrence patterns, which overlooks the influence of user intent. In fact, the intents of the users even with similar historical behaviors might be different and consequently need different complements. Second, the existing works often encode the complementary relationship at item level. In real world, however, different aspects of an item might contribute different complementarities to the same item, and the complementary information at aspect-level tends to be related with the intents of users. To overcome the two shortcomings, in this paper we propose a novel model called Aspect-level Complementarity Learning for Intent-aware Complementary Recommendation (AICRec). In particular, we propose a User Intent Perceiving (UIP) module, which enables AICRec to differentiate users' separate intents even though they are in similar scenarios. Meanwhile, we also devise an Aspect-level Complementarity Learning (ACL) module to infer an item's finer-grained complementarities to a user's intent at aspectlevel, which helps AICRec personalize the recommended complementary items with respect to the user's intent. At last, extensive experiments conducted on real datasets verify the superiority of AICRec over the state-of-the-art methods for complementary recommendation.(c) 2022 Elsevier B.V. All rights reserved.

Keywords:

Intent -aware recommendation

Complementary recommendation

Representation learning