Recommendation of what-to-buy

Abstract

Recommender systems has always been a hot research topic due to its prevalent usage in the ever-blooming e-commerce business. The exponential growth of available choices in e-commerce websites has brought about the information overload problem. With the help of recommender systems, high quality and personalized recommendations are provided to the users which help them easily locate items that match their preferences among numerous online products. This project intends to study the effectiveness of various recommender techniques in a real-world business setting and visualize the recommendation accuracy obtained from user feedback. To serve this purpose, an e-commerce website is developed through the course of the project using Django Oscar framework. Four recommendation algorithms, namely MostPop, UserKNN, PMF, and ReMF, are incorporated into the website. User ratings fetched from database are fed into the algorithms and recommendation results based on calculated prediction scores are displayed below the product catalogue for user reference. Feedback buttons are also implemented to register user feedback on the accuracy of recommendations. These user feedbacks are retained in the database and are used as inputs to calculate the recommendation accuracies for each algorithm. The results are visualized in a multi-bar chart to be displayed at the bottom of the catalogue page. The multi-bar chart always reflects the most up-to-date accuracy values to aid the users’ understanding of performance differences for different algorithms.