ATTENTIVE ITEM2VEC MACHINE LEARNING METHOD FOR RECOMMENDING TOURIST DESTINATIONS IN INDONESIA

Abstract

In today's digital era, recommendation systems have been used by various age groups. One of the recommendation systems that many people often use is the YouTube application, and users usually get video recommendations related to previously watched videos. This is evidence that this recommendation system is close and helps human life today. This makes researchers compete to develop better recommendation systems. One of the recommendation systems that is considered to have good performance is the Attentive Item2Vec (AI2V) method. The AI2V method is a development of the Item2Vec (I2V) method, which combines collaborative filtering with a neural network to get a recommendation. The advantages of AI2V are that it pays attention to the entire sequence of items (videos, images, etc) consumed by the user, the complex relationships between all items in the user's sequence, and can be applied in various recommendation systems. In this study, the AI2V method was applied to tourist assessment data on several tourist destinations in Indonesia (Jakarta, Yogyakarta, Bandung, Semarang and Surabaya). The data consists of 10,000 assessments given by 300 tourists. The AI2V methods include data preparation, data splitting, Attentive Context-Target Representation, Multi-Attentive User Representation, AI2V similarity function, Top 5 Recommendations, and performance evaluations. Based on the analysis used, AI2V produces the Top 5 Recommendations for tourist destinations, namely Keraton Surabaya, Desa Wisata Gamplong, Sanghyang Heuleut, Jogja Bay Pirates Adventure Waterpark, and Jogja Exotarium. The accuracy level with the Mean Percentage Ranking (MPR) is 0.48976, meaning that the recommended results for all tourists are considered reasonably good performance. The parameters used in the study include a learning rate (α) of 0.5, a window size of 500, and a max epoch of 50. The parameters that have been determined and the process run are expected to provide accurate recommendations and match the interests of tourists.