Knowledge-aware Dual-side Attribute-enhanced Recommendation

TL;DR

KDAR addresses the challenge of modeling fine-grained user preferences in knowledge-aware recommendation by extracting user preference representations from item attributes in Knowledge Graphs and pairing them with attribute fusion representations. It introduces a dual-side augmentation and a multi-level collaborative alignment contrasting mechanism to weight attributes according to Collaborative Filtering signals, combining attribute- and KG-based information for prediction. Empirical results on four benchmarks show KDAR consistently outperforms state-of-the-art baselines, with strong gains in Recall@20 and robust performance across settings such as cold-start and long-tail item recommendations. The approach highlights the practical value of leveraging the preference-attribute connection to enhance recommender systems in diverse domains, and provides code for reproduction.

Abstract

\textit{Knowledge-aware} recommendation methods (KGR) based on \textit{graph neural networks} (GNNs) and \textit{contrastive learning} (CL) have achieved promising performance. However, they fall short in modeling fine-grained user preferences and further fail to leverage the \textit{preference-attribute connection} to make predictions, leading to sub-optimal performance. To address the issue, we propose a method named \textit{\textbf{K}nowledge-aware \textbf{D}ual-side \textbf{A}ttribute-enhanced \textbf{R}ecommendation} (KDAR). Specifically, we build \textit{user preference representations} and \textit{attribute fusion representations} upon the attribute information in knowledge graphs, which are utilized to enhance \textit{collaborative filtering} (CF) based user and item representations, respectively. To discriminate the contribution of each attribute in these two types of attribute-based representations, a \textit{multi-level collaborative alignment contrasting} mechanism is proposed to align the importance of attributes with CF signals. Experimental results on four benchmark datasets demonstrate the superiority of KDAR over several state-of-the-art baselines. Further analyses verify the effectiveness of our method. The code of KDAR is released at: \href{https://github.com/TJTP/KDAR}{https://github.com/TJTP/KDAR}.

Figures (5)

• Figure 1: An example demonstrating the preference-attribute connection. In the figure, there are two users' historical interactions and preferences and a mini KG.
• Figure 2: The overall framework of KDAR. The graph aggregation is formed on CG and KG respectively to obtain aggregated representations. The Attribute-based representations are constructed with multi-level collaborative alignment contrasting. Finally, dual-side attribute level enhancement is performed and the enhanced representations are concatenated with KG-based representations to make predictions.
• Figure 3: The curves of Recall@$K$ on four datasets.
• Figure 4: The curves of Recall@$20$ and NDCG@$20$ when the number of GNN layers $L$ vary from 1 to 4 on four datasets.
• Figure 5: The Recall@$20$ on cold-start setting (\ref{['fig:cold_lfms']} and \ref{['fig:cold_yelp']})and long-tail item recommendation (\ref{['fig:long_lfms']} and \ref{['fig:long_yelp']}) on Last.FM and Yelp2018 datasets.