Dynamic Graph Collaborative Filtering

Abstract

Collaborative Filtering (CF) signals are crucial for a Recommender System (RS) model to learn user and item embeddings. High-order information can alleviate the cold-start issue of CF-based methods, which is modelled through propagating the information over the user-item bipartite graph. Recent Graph Neural Networks (GNNs) propose to stack multiple aggregation layers to propagate high-order signals. However, the oscillation problem, varying locality of bipartite graph, and the fix propagation pattern spoil the ability of multi-layer structure to propagate information. The oscillation problem results from the bipartite structure, as the information from users only propagates to items. Besides oscillation problem, varying locality suggests the density of nodes should be considered in the propagation process. Moreover, the layer-fixed propagation pattern introduces redundant information between layers. In order to tackle these problems, we propose a new RS model, named as Deoscillated Graph Collaborative Filtering (DGCF). We introduce crosshop propagation layers in it to break the bipartite propagating structure, thus resolving the oscillation problem. Additionally, we design innovative locality-adaptive layers which adaptively propagate information. Stacking multiple crosshop propagation layers and locality layers constitutes the DGCF model, which models high-order CF signals adaptively to the locality of nodes and layers. Extensive experiments on real-world datasets show the effectiveness of DGCF. Detailed analyses indicate that DGCF solves oscillation problem, adaptively learns local factor, and has layer-wise propagation pattern. Our code is available online 1 . * Equal contributions Preprint. Under review 1 https://github.com/JimLiu96/DeosciRec

Rank List Sensitivity of Recommender Systems to Interaction Perturbations

“…• Wikipedia [3, 14,33,38,44] contains the edit records of Wikipedia pages represented as (user, page, timestamp).…”

“…• Reddit [2,14,33,38,44] includes the posting history of users on subreddits represented as (user, subreddit, timestamp).…”

“…We emphasize that the datasets are diverse, standard, and popular in the recommendation community. LastFM is a widely used recommendation benchmark dataset [22,29,35,47], Foursquare is broadly utilized for point-of-interest recommendation tasks [62,[64][65][66], and Wikipedia and Reddit are popular for social network recommendation tasks [14,33,38,44]. We select these datasets for experiments because (a) they come from diverse domains, thus ensuring generalizability; (b) the timestamps of interactions reflect when the corresponding activities happened (as opposed to Amazon review datasets where a review is posted much after a product is purchased, or MovieLens review dataset where a review is posted much after a movie is watched); and (c) these are popularly used datasets in numerous papers as listed above.…”

Fairness Embedded Adaptive Recommender System: A Conceptual Framework

“…Latest advances in the field include using graph encoding, Stochastic Shared Embeddings, large-scale Pairwise Collaborative Ranking, Sequential Recommendation Via Personalized Transformer [43] that mainly solve the problem of scaling to massive datasets, learn user and item embeddings and think about the problem as a sequence of actions, not oneshot recommendations. The user and items embeddings are mainly a different way to refer to latent variables and a series of work leverage the power of Neural Networks to try and learn them [44][45][46].…”

Relation-aware Heterogeneous Graph for User Profiling

“…In user profiling, an intuitive way is to model the user's interaction behaviour with graphs. Despite the success of traditional deep learning approaches [3,9,13], graph methods are highlighting their advantages on non-euclidean relations in such tasks [7,18,[20][21][22][23][24]. Chen et al [2], Rahimi et al [12], Xiao et al [19] regard users with co-relation (like co-purchase in e-commerce) as a graph with entities and hierarchically pump the heterogeneous information up from the attribute with graph attention networks.…”