A method for classifying commodity information with the fusion of attribute features

In this paper, we propose a deep learning-based method for classifying commodity information to improve the problem that the model needs to understand the semantics of commodity information sufficiently, which leads to a reduced classification effect when the commodity descriptions are similar and complex. The research context is set to describe commodity information containing precious metals in the wording. The method enhances the commodity classification effect by extracting attribute entities in the commodity information, including identification attributes, commodity descriptions, and commodity specifications, and using a self-attention mechanism to fuse the attribute entity features with the overall semantic features to highlight the category features in the commodity information. The experiments show that the method improves the F1 value and accuracy rate by 2.28% and 1.81%, respectively, compared with typical classification methods, proving that fusing the attribute features of commodities can improve the semantic understanding of the model for different kinds of commodities under similar descriptions, which has some practical application value for dealing with complex semantic problems in commodity information.