Online Product Recommendation using Relationships and Demographic Data on Social Networks

Objectives: There have been many traditional product recommender systems in the past, but they were mostly based on collaborative or content filtering, historical transaction records or website browsing history of the users. This approach leads to sparsity and cold-start issues. Methods/Statistical Analysis: The proposed system is a hybrid one and can improve the suitability and the accuracy of recommender systems with the help of users’ and products’ demographic info and the ratings of a brand name, constantly updated in the e-commerce and social media websites. Findings: Experiments on the actual dataset reveal that friends with similar tendencies select the similar items and this approach can solve the relevant data sparsity and cold start problems. Initially, the system loads the data about the friends who have given product ratings and selects a target cold start user. Then it predicts his/her tech level and cost level using BMART algorithm. With the help of these, it finds the user-friends and user-items similarity using the cosine and Euclidean cluster similarity methods respectively. It then feeds these into the proposed ranking algorithm to find the relevance scores of the products to the target user and based on this the recommended product links are displayed. Application/Improvement: The outputs reveal that the proposed system gives very accurate and personalized product recommendations to the user.