CoupledCF: Learning Explicit and Implicit User-item Couplings in Recommendation for Deep Collaborative Filtering

Abstract

Non-IID recommender system discloses the nature of recommendation and has shown its potential in improving recommendation quality and addressing issues such as sparsity and cold start. It leverages existing work that usually treats users/items as independent while ignoring the rich couplings within and between users and items, leading to limited performance improvement. In reality, users/items are related with various couplings existing within and between users and items, which may better explain how and why a user has personalized preference on an item. This work builds on non-IID learning to propose a neural user-item coupling learning for collaborative filtering, called CoupledCF. CoupledCF jointly learns explicit and implicit couplings within/between users and items w.r.t. user/item attributes and deep features for deep CF recommendation. Empirical results on two real-world large datasets show that CoupledCF significantly outperforms two latest neural recommenders: neural matrix factorization and Google's Wide&Deep network.

Critically Examining the Claimed Value of Convolutions over User-Item Embedding Maps for Recommender Systems

“…Learnable embeddings: Also in this case the CNN is applied on user or item embeddings. Differently from the previous case, the embeddings are an integral part of the model and are trained along with the CNN (i.e., they are not pre-trained with another approach), e.g., in [17,48].…”

“…However, the papers analyzed in this work, i.e., [9,13,43,48], do not clearly provide a definition of locality for the embeddings, nor do they describe the semantic topology of the input data. In the ConvNCF approach [13], for example, the input to the CNN is a user-item interaction map that is created by computing the outer product of embeddings pretrained using matrix factorization.…”

“…Coupled Collaborative Filtering (CoupledCF) was proposed in [48] as a method to learn implicit and explicit couplings between users and items, taking advantage of them not being independent, to leverage side information more effectively (e.g., user demographics, item features). The model is composed of two cooperating architectures, one is a deep collaborative filtering model which only uses user-item interactions, while the other is a CNN on user and item embeddings whose aim is to learn the couplings.…”

“…In some recent works, different proposals were also put forward to combine the proven power of low dimensional approximation models (e.g., latent factor models) with CNNs. In particular, one underlying idea of the proposals in [9,13,43,48] is to use CNNs to model and leverage correlations in the latent factors (embeddings) space. In all these papers, the respective authors claim to have obtained significant gains in accuracy by applying convolutions over user-item interaction maps derived from the outer product of user-item embeddings.…”

Evolutionarily Learning Multi-Aspect Interactions and Influences from Network Structure and Node Content

“…It is important for us to understand the multi-aspect node influences and the multiaspect interactions between nodes that drive the formation of connections. The above example illustrates a critical perspective in complex networks and systems, i.e., representing multiaspect and heterogeneous interactions and influences between nodes (objects), in order to understand some intrinsic characteristics and fundamental complexities: explicit and implicit heterogeneities and coupling relations in complex networks and systems (Cao 2014;2015;Zhang et al 2018a). This requires to involve, model and integrate (1) explicit and heterogeneous sources of node content information (e.g., paper's title and/or abstract) and network topological structure (e.g., paper citations), and (2) implicit and heterogeneous aspects of node influences (e.g., a paper's topic uniqueness and design novelty) and node interactions (e.g., a paper cites the algorithm introduced in another paper).…”