Recommender engine using cosine similarity based on alternating least square-weight regularization

By the growth of digital data which leads to more complex demands from user to find the information or items. Search engines solve most of the problems but have the drawback, it depends on the query/term that the user enter. The problem appears when the user forget or does not know the query that associated with the items. The Recommendation comes as a solution to provide personal information by studying the interaction of a user, user community, and items that have been recorded previously. Collaborative filtering as a method to provide personalized recommendations based on other users who have similar tastes. However, the results of collaborative filtering tend random, sometimes users need an item with similar genre/subjects. This paper discusses a model of a recommendation engine for new users with a method of collaborative filtering based on genre similarly with the aim of giving the smallest error with high precision. First filter we use Alternating Least Square-Weight Regularization (ALS-WR) is selected as algorithms for collaborative filtering. Second filter we use Cosine Similarity is selected as an algorithm for genre similarity. We use datasets from movielens.org. The RMSE on the first recommendation generated is 0.89 for 100K ratings, 0.86 for the 1M ratings, and 0.81 for the 10M rating. By iterative and training on larger data, it will make a better model, so RMSE can be smaller. They are concluded that ALS-WR able to deliver adaptive, with regulatory parameters that can be controlled and adjusted. The more data but the error on the wane, that is means this algorithm is suitable for growing data or big data. The item that has been sorted with the ALS-WR algorithm, letter approximated with cosine similarity, and with only 10 items movie displays with the highest degree of similarity, that be able to generate high precision.