Özet:

Location-based social networks have evolved significantly over the past decade by providing us with a new platform to investigate their preferences based on the user’s location history. Most location-based social networks provide a variety of spaces placed under a category hierarchy in which users can explain, interpret or leave clues. Although geographically informed location recommendations have attracted the interest of many researchers, the impact of the research projects on the user’s preferences has been ignored in most. Since a user can choose different places to visit at different hours of the day, two users who enter the same amount in a particular category may be less similar depending on the time they are in that place. Furthermore, traditional collaboration-based filtering techniques take into account the preferences of all users, but only taking into account the preferences of the category specialists, recommending a place in that category can further enhance the recommendation performance. For this purpose, a new scalable time-based, expert-backed collaboration-based filtering approach was proposed to recommend spaces. In this new approach, first and foremost, the expertise of users according to categories is examined, taking into account the diversity of entry records made in a particular category, and the top m of the experts for each category is selected. Then, the preferred score for each user-interface pair is calculated in relation to the time range of the day. Three algorithms have been developed to take into account the entry recording time information in different stages. The first algorithm calculates the similarity values between users for each predefined time range. The second algorithm takes into account the user-mekan input frequency matrix in this specific time range. The third algorithm benefits both the similarity values and the user-medium input frequency matrix of the time range. Finally, the top-k following locations are presented as a recommendation to the user. The suggested algorithms are evaluated with two large-scale Foursquare data sets and compared with the basic approaches.

Anahtar Kelimeler: