H&M Recommendations

Recommendation systems have a crucial role in online user services. They are used in many industries such as music, movies, and shopping, ensuring customer satisfaction and accelerating sales. With the increase in fashion trends globally, various options are available on e-commerce websites available to the customers. Selecting appropriate clothing options has become a challenging task, given the many choices available online. Unsatisfied customers can also lead to fewer returns/profits for the company.

Our focus for this project was on the H&M clothing company and used a dataset from Kaggle. We were able to develop a deep neural network retrieval model using TensorFlow Recommenders to generate recommendations for users that visit H&M online. The retrieval models were trained using the two-tower architecture for the Query(User) and Candidate(Product) models. Our contextualized models were successfully trained using Horovod for distributed training and achieved a 94% Top-5 accuracy on our test set.