As our catalog of products grows, suggesting items that users are likely to like is essential to keep our users enthusiasm high. There are several levers upon which we can act to improve our user activation, one of them being the use of a recommender system to reduce users eye friction. Recommendation systems are a class of algorithms that allow to make suggestions in a personalized manner, based on users past activities. In our specific case, we are interested in developing an image-based recommendation system capable to predict whether a user will like a product or not, depending on its attributes.

In order to achieve this objective, it is required to build an adequate set of clothing features, so that each product can be decomposed into relevant features that may help us understand our users preferences better. As for the model, the intern will start with a regularized matrix factorization to build the recommender system and will be asked to solve the associated optimization problem using different approaches (Alternating Least Squares, Stochastic Gradient Descent) to help us decide which method should be implemented, depending on their results consistency and computational speed. If time allows, the intern may also:

• help in the construction of an image dataset specific to fashion

• integrate a Convolutional Neural Network to automate the process of feature extraction

• compare its performance against a classical matrix factorization with features manually extracted

For students in: Machine Learning, Statistics, Applied Mathematics

Requirements

• willingness to learn
• problem-solving capabilities
• proficiency in Python (pandas/scikit-learn/TensorFlow) or R
• Machine Learning background
• knowledge in Computer Vision and Convolutional Neural Networks is a plus
• basic knowledge of SQL is preferable but not mandatory