Recommendation algorithm for agricultural products based on attention factor decomposer and knowledge graph

Abstract

To alleviate the distress of data sparsity and cold start in agricultural products e-commerce platforms, this paper proposes an agricultural products recommendation algorithm based on the combination of attention factor decomposer and knowledge graph. The algorithm constructs a knowledge graph for the produce dataset, models the higher-order connectivity of the produce knowledge graph in an end-to-end manner under the space of the knowledge graph, recursively propagates embeddings from the neighbors of the nodes, and extracts the potential feature vectors of the produce by using the attention factor decomposer as the message aggregation of the neighboring nodes. Using MLP, the agricultural product feature vectors and user embedding vectors are integrated and sent to the prediction module, and user click-through rate prediction is obtained by vector inner product operation. Experimenting on an agricultural e-commerce dataset, the ACC and AUC are improved by 1.60% and 1.14%, respectively, compared with the optimal baseline model KGCN. Thus, it verifies the effectiveness as well as feasibility of the improved algorithm on agricultural products data, which can provide a new idea and method for agricultural products e-commerce platform.