Inferring Representations for Cold-Start Recommendations

Abstract

This thesis investigates the development of an effective hotel recommendation system by leveraging neural networks and advanced modeling techniques. The present research aims to comprehend the difficulties related to recommendation systems, specifically collaborative filtering and the cold start issue. The study investigates deep and recurrent neural networks and their training procedures to develop strong model architectures that enhance recommendations. The experiments involve processing datasets, performing analytics, and developing various models such as Deep Average Network (DAN), and sequential model variations. Innovative solutions, such as embedding enhancement and a hotel embeddings calculator, are proposed to tackle the cold start problem. The study shows that the suggested method is efficient in improving hotel recommendation systems. The thesis enhances hotel recommendation systems by presenting ideas and techniques that can be further developed and improved in future research.