Mitigating Cold Start Problem in Recommendation Systems via Transfer Learning Approach

Abstract

Recommender system provides personalized services to its customers from a huge amount of choices available to them in different domains i.e. E-commerce, Music streaming apps, E- Learning platforms. Understanding the preferences of new users is a challenging task in recommendation systems, commonly known as Cold Start problem, since no prior knowledge about the users interests is provided. In such scenarios, data from social networking sites could be utilized to determine user preferences. In this work, we have proposed a novel Dove Regression based Recommendation System (DRbRS) which employs a Transfer learning approach that utilizes user interactions from social media sites and applies learnt information to suggest pertinent products to new users. The user profile is built using information directly provided by the user and predictions are made based on this classification using machine learning methods like classification trees. The results and comparative analysis of the proposed model indicate better performance and various performance measurements, including accuracy, precision, Recall, and F1-score were evaluated.