Online and mobile application shopping has become common place for consumers. They have access to numerous e-Commerce sites that offer consumers an even larger selection of consumer products than what can be found in traditional offline retail and an entirely different way to shop compared to traditional offline shopping. Consumers can use their personal preferences (size and color) to filter a large selection of items down to the products they want and then compare them with multiple sites before deciding what product to buy. This research models the user to item relationships through a confidence weighted version of Alternating Least Squares (ALS) trained using implicit user feedback. In parallel to this, I will encode the metadata of items using TF-IDF and then reduce these representations using a truncated Singular Value Decomposition (SVD) to obtain a representation of semantic similarity between items. The final ranking of items will be based on the weighted fusion of the collaboration and content scores, where the model training occurs offline and the inference from this model will occur through an in-memory REST service for low latency response time.
