Use of Hybrid Methods in Making E-commerce Product Recommendation Systems to Overcome Cold Start Problems

Abstract

The large number of users and the items offered in e-commerce make it difficult for buyers to choose the right items and sellers to offer their items to the right buyers. To overcome this problem, a system that can offer and recommend goods automatically, namely a recommendation system is needed. One of the most popular methods used to create a recommendation system is collaborative filtering, the recommendations are created based on similarities in user behavior. Unfortunately, this method has a weakness, namely cold start, where the recommendations will be inaccurate on data that has a lot of new users and items due to minimal historical data regarding user behavior. This problem will be tried to be solved in this study using a hybrid method, where this method combines more than 1 method to create a list of recommendations so that it will cover the shortcomings of each method. This study uses Amazon's e-commerce product and transaction data. The use of the hybrid method in this study can overcome the cold start problem by using switching and mixed methods, by not using the collaborative filtering model on new user recommendations or users who have little interaction. New users will receive recommendations based on the combination of popularity-based and content-based filtering models. This can be seen from the Mean Absolute Error (MAE) value of the model, where the MAE value for the data with a minimum user has at least 3 times rating is 0.566883, for the minimum 7 times, the MAE value is smaller, 0.487553.