Maximum Coverage Method Modification with Timeliness in Non-Personalized Recommendation for Pure Cold-Start Users
Novan Alkaf Bahraini Saputra, Wikan Danar Sunindyo · 2019
A recommendation system should be able to provide recommendations to early users who are referred to as pure cold-start because it is related to the main business KPI, which is giving initial impressions to the initial user and increasing the percentage of initial users to be users of the system. To overcome this, the non-personalized approach is applied by using the maximum coverage method to cover users as much as possible and timeliness as an aspect of time to be able to make different items that have higher relevance for the initial user get to the recommendation. The study was carried out on the 100k and 1M movielens datasets using maximum coverage timeliness in 2 scenarios, namely scenario 1 using all item data as input while scenario 2 cutting the number of items based on user coverage of the item. Resulting in the dataset movielens 100k, experiments with the name MaxCovTL_100 in scenario 2 have the best utility in top-15 and top-20 and the best covered users in top-15 compared to other methods. But the improvement was not significant after testing with the Wilcoxon test. In the 1M dataset movielens, the results of recommendations from the maximum coverage timeliness method are very similar to the maximum coverage. so that this experiment can be said to be unsuccessful.