Enhancing Product Recommendation Accuracy Using Bipartite Link Prediction and Long Short-Term Memory in Retail Industry

As competition in the retail sector intensifies, the demand for accurate customer-product recommendation systems has grown. Traditional similarity-based approaches such as common neighbor, Jaccard, Adamic Adar, preferential attachment, and resource allocation have been widely adopted in many business applications. However, these methods often struggle with capturing complex purchasing behaviors, product heterogeneity, temporal demand variations, and scalability challenges. This study introduces a deep learning-based recommendation model that integrates bipartite link prediction networks with Long Short-Term Memory (LSTM) to improve predictive accuracy. The bipartite network represents customer-product interactions, while the LSTM model captures sequential purchasing patterns to forecast future transactions. Experimental evaluation on a real-world building materials retail dataset comprising 389,087 transactions demonstrates the effectiveness of the proposed approach, achieving a Precision of 0.8223, Recall of 0.8034, F1-score of 0.8128, NDCG of 0.8601, and overall prediction accuracy of 0.854. The results indicate that the proposed model significantly outperforms similarity-based techniques, offering a robust solution for enhancing recommendation performance in dynamic retail environments.