Personalized User Recommendation based on Various User Behavior in Local Domain
Junhyung Kim, Yeonghwan Jeon · 2022 IEEE International Conference on Data Mining Workshops (ICDMW) · 2022
User recommendation is a service that recommends other users to a user using the Social Network Service (SNS). This user recommendation is important because it can make user's active behavior in service. Generally, this user recommendation has been implemented based on similarity between user node and user node in structure of network (graph) or a simple user profile. However, this method does not reflect user's taste. In this paper, we propose an empirical user recommendation problem which reflects user's taste in local domain. Unlike other domains, user recommendation in local domain should work in consideration of various components such as user, region, Point-of-Interest (POI), and User Generated Content (UGC). To deal with the problem, we introduce how to define users with similar taste by various user behavior and how to make user embedding by neural network according to these definitions. Furthermore, since there is no right answer in recommendation, we suggest a method of showing optimal result to user by increasing diversity and novelty using ensemble, in the initial stage for multi-armed bandit.