Delhaize, a leading Belgian supermarket chain, successfully enhanced its online customer experience by implementing a refined recommendation algorithm for out-of-stock items. Partnering with us, Delhaize aimed to optimize the algorithm’s performance and expand its capabilities.
Utilizing product descriptions and internal hierarchies, we elevated the model by incorporating textual and hierarchical similarity components. Textual similarity was achieved through TF-IDF-based word embeddings, while hierarchical ordering used both numerical and textual representations, employing log difference and Jaccard similarity calculations.
To create a robust solution, we integrated these new components with the existing algorithm, forming an ensemble model. This design allowed for easy experimentation with component weights and ensured straightforward future model updates. The model was industrialized using Azure, enabling scheduled reruns and continuous, automatically updated recommendations.
- A more accurate recommendation algorithm
- Streamlined experimentation for model optimization
- Automated, up-to-date recommendations, enhancing customer satisfaction and engagement.
Technologies employed: Azure, Databricks, SQL, Pyspark, Python