Design of Friends-Making Recommendation System Based on Reinforcement Learning and Generative Adversarial Network

Abstract

With the popularity of mobile phones, people are more interested in searching of friends through the dating applications. In the dating applications (APP), people are distressed by the vast amount of data and junk information, so the dating system of recommendation came into being. General dating software adopts content-based recommendation algorithm and collaborative filtering recommendation algorithm, but it has some problems, such as being unable to mine users' potential preferences and difficult to extract features. In recent years, users want to search for friends through face images. it is an urgent problem to recommend their interested friends to users, according to users' preferences. This paper designs a dating recommendation system based on reinforcement learning and generative adversarial network (GAN), Including its scheme design and programming implementation. The main flow is as follows: firstly, GAN generates false faces and matches real faces to users through false faces to face database. Secondly, users give the evaluation feedback on faces. Thirdly, the system optimizes the false faces generated by GAN through reinforcement learning according to the user feedback. Finally, it iterates continuously until users are satisfied. The implementation of the system is mainly divided into two parts, one is GAN face generation, and the other is the strategy of optimizing GAN generation by reinforcement learning.