Cosine Similarity Measurement for Indonesian Publication Recommender System

Abstract

So many publications are increasing every year and it causes an overflow of data and it makes the information function on the site cannot be delivered entirely to the visitors. As a scientific publication site, the IPI garuda portal (Index of Indonesian publications) has more than 4000 Indonesian journals in the database. Therefore it will be designed a system that can provide information and knowledge for the user. The system can also provide recommendations related articles and relevant for users. Recommendations are made by calculating similarities between related documents. The similarity calculation method used in this research was cosine similarity. The steps taken were pre-processed, weight calculation, vector length and then they were calculated using cosine similarity so that the value would be achieved that ranged from 0 to 1. Performance testing of recommendation system used precision and recall. Performance system with average precision value about 0.664 meant that the document had good system accuracy, and then the average recall point with value about 0.962 meant that documents were successfully returned by the system.