A hybrid recommendation system using association rule mining, i-ALS algorithm, and SVD++ approach: A case study of a B2B company

In the field of recommendation systems, collaborative filtering is a widely used technique. It provides recommendations to active users based on the ratings provided by similar users. However, this method may reduce the accuracy of user preference predictions and lead to lower-quality recommendations in cases of high data sparsity. This issue is often observed in the Business-to-Business (B2B) context, where user-generated reviews are often sparse. To overcome this challenge, we present a novel hybrid approach that explores product taxonomies and association rule mining combined with an advanced method for initialization. Our approach first involves generating a new explicit taxonomy based solely on textual product descriptions and extending the user-product matrix using association rule mining results. Second, complementary items are added to the user-item matrix based on users’ purchasing behaviors, as emphasized by the extracted association rules. Finally, we use the implicit Alternating Least Squares (i-ALS) algorithm and initialize the latent factor matrices with values obtained through the singular value decomposition approach (BLS-SVD++). This hybrid approach is tested and compared with conventional approaches, considering a real-world case study of a distributor located in Quebec. The results obtained from feedback implicitly inferred from sales data demonstrated improved RS performance compared to conventional approaches. • A novel hybrid recommendation system is proposed. • We used association rule mining, the i-ALS algorithm, and the SVD++ approach. • This hybrid system is particularly relevant for overcoming the sparsity problem in B2B settings. • Real-world B2B sales transactions are considered. • The proposed methodology can be generalized to other real-world cases.