Intelligent personalized shopping recommendation using clustering and supervised machine learning algorithms.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Next basket recommendation is crucial for market basket analysis, especially in grocery shopping.
• Grocery lists are integral to consumer shopping habits.
• Existing systems lack personalization for individual customer needs.

Purpose of the Study:

• To introduce a novel grocery recommender system on the MyGroceryTour platform.
• To provide personalized weekly grocery lists for Canadian customers.
• To compare traditional machine learning (ML) and deep learning (DL) algorithms for recommendation accuracy.

Main Methods:

• Clustering analysis to segment customer shopping habits into four clusters.
• Implementation of traditional ML algorithms, including Random Forest.
• Development of a new DL algorithm using a gated recurrent unit (GRU)-based recurrent neural network (RNN) architecture, extending the DREAM model for multi-class classification.

Main Results:

• The proposed DL algorithm achieved an average F-score of 0.559, outperforming Random Forest's F-score of 0.516 on a dataset of 831 customers.
• The system offers real-time recommendations by integrating purchase history, store specials, and product availability.
• Personalized models built for each customer significantly improved prediction accuracy over general DL models.

Conclusions:

• The developed recommender system effectively generates personalized intelligent grocery lists.
• The DL algorithm, an extension of DREAM, demonstrates superior performance in multi-store recommendation tasks.
• Personalization at the individual customer level is key to enhancing the accuracy of next basket recommendations.