Porat^Rank to Improve Performance Recommendation System

The e-commerce recommendation system has experienced tremendous progress and has caused an explosion of information, making it difficult for users to choose items that fit their preferences and take a long time. One way to overcome this condition is to use a collaborative filtering approach. Collaborative filtering generally uses similarity measurements and ranking predictions to produce recommendations. However, the recommendations presented are less reliable when data conditions are sparse. This condition encourages the development of ranking-based collaborative filtering. Some ranking-based methods are Copeland and Borda, which carry out an aggregation process to produce product ratings that are recommended to users. Both of these methods use limited ranking data at the user preference profile stage and do not involve re-ranking data during the aggregation process. This process causes the resulting recommendations to decrease in quality. Therefore, this paper proposes the Porat^Rank method. The basic idea of this method is to optimize the utilization of ranking data to produce product ratings that are more in line with user preferences. Ranking data is used as an additional factor in determining product points. Determination of product points not only looks at the ranking value but also considers the same number of ratings, and the position of the product in its appearance. It also sees the effect of the ranking value using the minus function. Optimizing ranking data in the aggregation process can improve the recommendation results, as shown by the average value (NDCG) of the Porat^Rank method, which is higher than the Borda and Copeland methods. The Porat^Rank method is faster than the Copeland method and manages to overcome the problem of sparsity and scalability, which is a major problem in the collaborative filtering approach, so the Porat^Rank method is feasible to be used in improving performance recommendations system.