Generalizable Potential Supplier Recommendation Under Small-Sized Datasets via Adaptive Feature Perception Model

Precisely deciding potential suppliers enables companies to engage with high-caliber partners that fulfill their strategic development requirements, bolster their core competitiveness, and foster sustainable market growth. To mitigate the challenges enterprises face in selecting appropriate suppliers, a recommendation method for potential suppliers tailored to a small-sized dataset is proposed. This approach employs an enhanced Graph Convolutional Neural Network (GCNN) to resolve the accuracy deficiencies in supplier recommendations within a limited dataset. Initially, a supply preference network is created to ascertain the topological relationship between the company and its suppliers. Subsequently, the GCNN is enhanced through dual-path refinements in network structure and loss function, culminating in the adaptive feature perception model. Thereafter, the adaptive feature perception model is employed to adaptively learn the topological relationship and extract the company’s procurement preference vector from the trained model. A matching approach is employed to produce a recommended supplier list for the company. A case study involving 143 publicly listed companies is presented, revealing that the proposed method markedly enhances the accuracy of potential supplier recommendations on a small-sized dataset, thereby offering a dependable and efficient approach for enterprises to effectively evaluate potential suppliers with limited data.