A hybrid approach in food recommendation: Challenges and implementation

With the steady increase in the use of online platforms, ranging from social media to streaming services, it becomes essential to personalize user experiences in order to induce continued use. In this context, recommender systems emerge and evolve, leveraging user and object data to increase user engagement. The central objective of this thesis is the design and implementation of a custom recommendation system for the Weeshop application, leveraging data collected from this platform. Initially, a hybrid system integrating content-based, collaborative-based and session-based parts was conceived. However, due to the scarcity of user data, the implementation of the collaborative component was discarded. As for the content-based component, a neighborhood-based approach was adopted, which was also employed for the secondary objective of classifying new products to be added to the item database. The performance of this algorithm was compared with that of a k-Means type clustering algorithm, but the results indicate better performance using the former approach. As for the session-based part, we implemented a recurrent neural network using LSTM cells by explaining the choice of hyperparameters step by step. However, the results obtained were not satisfactory, and consequently we excluded this component from the system.