Developing recommender systems for personalized email with big data

Abstract

Recommender systems are nowadays widely used in e-commerce industry to boost its sale. One of the popular algorithms in recommender systems is collaborative filtering. The fundamental assumption behind this algorithm is that other users' opinions can be filtered and accumulated in such a way as to provide a plausible prediction of the target user's preference. In this paper, we would like to develop a recommender system with big data of one e-commerce company and deliver the recommendations through a personalized email. To address this problem, we propose user-based collaboration filter based on company dataset and employ several similarity functions: Euclidean distance, Cosine, Pearson correlation and Tanimoto coefficient. The experimental results show that: (i) user responses are positive to the given recommendations based on user perception survey (ii) Tanimoto coefficient with 10 neighbors shows the best performance in the RMSE, precision and recall evaluation based on groundtruth dataset.