Neural Causal Graph Collaborative Filtering

TL;DR

NCGCF addresses inaccurate embeddings in graph-based collaborative filtering by injecting explicit causal structure into graph learning. It constructs a Structural Causal Model for users, items, and preferences and parameterizes it with neural networks to form a Neural Causal Model, trained via variational inference and a counterfactual ELBO. A semi-implicit causal graph encoder and causality-aware message passing produce embeddings that reflect true causal influences, yielding superior Precision, Recall, and NDCG across synthetic and real datasets. This causality-aware framework improves robustness to user-preference shifts and enhances explainability in graph-based recommendations.

Abstract

Graph collaborative filtering (GCF) has gained considerable attention in recommendation systems by leveraging graph learning techniques to enhance collaborative filtering (CF). One classical approach in GCF is to learn user and item embeddings with Graph Convolutional Network (GCN) and utilize these embeddings for CF models. However, existing GCN-based methods are insufficient in generating satisfactory embeddings for CF models. This is because they fail to model complex node dependencies and variable relation dependencies from a given graph, making the learned embeddings fragile to uncover the root causes of user interests. In this work, we propose to integrate causal modeling with the learning process of GCN-based GCF models, leveraging causality-aware graph embeddings to capture complex causal relations in recommendations. We complete the task by 1) Causal Graph conceptualization, 2) Neural Causal Model parameterization and 3) Variational inference for Neural Causal Model. Our Neural Causal Model, called Neural Causal Graph Collaborative Filtering (NCGCF), enables causal modeling for GCN-based GCF to facilitate accurate recommendations. Extensive experiments show that NCGCF provides precise recommendations that align with user preferences. We release our code and processed datasets at https://github.com/Chrystalii/CNGCF.

Figures (8)

• Figure 1: Paradigms of user preference modeling in a class of CFs: (a) Early CF, (b) GCF, and (c) Our causality-aware GCF. $Z_u$ represents hidden exogenous variables for users, e.g., user conformity; $Z_v$ are hidden exogenous variables for items, e.g., item exposure. $U$ and $V$ denote user and item, respectively. $E$ denotes preference representations from graph representation learning. $Y$ represents users' predicted recommendations.
• Figure 2: NCGCF framework: causal graph conceptualization prepossess a user-item interaction graph by using the causal relations under our defined Causal Graph; causal graph encoder models the causal relations under the graph-structured data using a semi-implicit generative model, and outputs user and item representations with a user encoder and an item encoder; collaborative filtering (CF) decoder uses CF to construct preference vectors based on user and item representations. Finally, NCGCF is optimized through a counterfactual instance-aware ELBO to capture user preference shifts.
• Figure 3: Impact of embedding size on Amazon-Beauty.
• Figure 4: Impact of embedding size on Amazon-Appliances.
• Figure 5: Impact of embedding size on Epinions.

Theorems & Definitions (4)

• Definition 2.1: Causal Graph
• Definition 2.2: Problem Formulation
• Definition 3.1: Structural Causal Model
• Definition 3.2: Neural-Causal Connection