User Profiling and Interest-based Machine Learning to Solve the Cold-Start Problem for a Recommendation Engine

The cold-start problem is a major issue in recommender systems that becomes a challenge for e-commerce companies when processing large amounts of data. Several strategies have been proposed to overcome the cold-start problem using different approaches, but the user interest approach is a unique and standalone approach. In this study, we present an interest-based approach that establishes a relationship between user profile data and corresponding user interests. Our work will recommend the items to the cold-start users based on their pro-files and interest matching, and also promote the cold-start users. Our approach is unique in that we consider user interests rather than similarities between items and recommend items which are based on user interests. The experiments and related tests show that proposed method is efficient and successful as far as of classification time, training time, and precision. Although other classifiers were efficient in terms of accuracy, the Bayesian classifier and Chain Classifier had the shortest classification time as well as training time.