Brand Positioning Analysis and Competitor Recommendation System Combining Knowledge Graph and Graph Attention Network
Year: 2026
Type: conference-paper
Abstract: Accurate competitor identification and brand positioning are critical for strategic decision-making in modern marketing. However, traditional recommendation methods, such as Collaborative Filtering (CF), suffer significantly from data sparsity in brand-user interactions and fail to capture the rich semantic nuances embedded in user-generated content. To address these challenges, this paper proposes a novel framework named KG-GAT (Knowledge Graph Attention Network) for competitor recommendation. We first construct a heterogeneous brand knowledge graph that integrates brands, products, and fine-grained attributes extracted from unstructured reviews. Subsequently, a Graph Attention Network (GAT) with a multi-head attention mechanism is employed to propagate high-order collaborative signals. This mechanism dynamically assigns importance weights to different neighbor nodes, effectively filtering out generic semantic noise while highlighting distinctive brand characteristics. Extensive experiments on a real-world dataset comprising 1,200 brands across eight industries demonstrate the superiority of the proposed approach. KG-GAT achieves a Precision@5 of 0.824, outperforming traditional CF and standard Graph Convolutional Networks (GCN) by 19.2% and 8.3%, respectively. Furthermore, the visualization of attention weights provides interpretability, revealing the specific attributes that define competitive relationships.
Authors: Wen Hui Zhong, Jian Zhang, Gelegjamts Adilbish
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204953058