Existing recommender systems like HSPRec19 and HPCRec18 improve recommendation accuracy but do not further capture item-level user multi-behavior interactions, such as clicks, likes, carts, and purchases. While DCABRec21 and MBHT22 attempt to use user-item multi-behavior interactions, they are not sequential recommendation systems. This paper proposes a Multi-behavior Sequential Pattern Recommendation System (MBSPRec), an extension of the HSPRec19 system, which incorporates users’ multi-behavior patterns to enhance the quality of the user-item rating matrix input to recommendation system, reducing matrix data sparsity, and thereby improving recommendation accuracy. MBSPRec creates a Multi-behavior Sequential Database for user behavior types, ApproxMap Algorithm mines frequent sequential rules for generation input rating data for new items without previous user interaction. Experimental results of MBSPRec on publicly available datasets in comparison with existing systems of HSPRec19, MBHT22, and DCABRec21 show improved recommendation accuracy in terms of Precision, Recall, and Mean Reciprocal Rank, therefore reducing user-item rating data sparsity.