A simplified method for improving the performance of product recommendation with sparse data

Abstract

As the development and population of Electronic Commerce (EC) keeps growing, the sheer volume of data makes the EC more challenge to handle. For example, if the number of products keeps increasing, a regular user-item transaction records may get bigger and bigger which will usually form the sparse matrix. When the data sparsity is formed, the performance of data analysis should be aware. In order to promote the product under hundreds or thousands of items, recommender systems have been intensively studied and are highly applied in the EC environment. One of the most popular recommendation methods is the collaborative filtering (CF) method in which a group of user/customer with similar preference is chosen for the reference of recommendation. It is interesting to notice that the past researches usually didn't consider the sparsity problem when applying the CF recommendation. As a result, the time performance of CF recommendation is, therefore, constrained. To overcome this problem, this research proposes a Simplified Similarity Measure (SSM) for CF recommendation when dealing with the sparsity problem. The proposed SSM method uses the real data from Epinion.com for experiment and for comparison. The results show that SSM method is outperformed the traditional CF methods in terms of time efficiency and recommendation precision.