A Novel Link Prediction Model Based on Deep Reinforcement Learning

Authors

  • Dawud, Jidda Jidda, Fengyu, Yang

Abstract

The Link prediction is considered, a significant piece of work in scientific collaboration
networks. Link prediction is a key fact-finding topic in the field of networks structured and graphs. Its
main objective is to detect pairs of nodes that will either form a connection or not forming a connection
in the future, with new research on prediction has been made by different scientist toward link
prediction. Although a lot of effort has been made by researchers to develop new prediction approaches,
currently more researchers have been doing their research on machine learning, recently research’s leads
to different way of link prediction with better and accurate prediction. Recently, owing to rapid
development of social-based online systems, coupled with the availability of huge data amount and
dynamic changes of networks, link prediction brings about challenges when trying to analyze the data.
Using deep re-enforcement learning mechanism, we put forward a novel link prediction methodology in
this research work. This proposed method introduced the intelligent retrieval method, for link prediction
on gregarious connections between nodes. In order to predict the advent of impending relations among
nodes pairs at important time-stamps which possess close relationships in network, this research then
focuses mainly on plan for linking prediction models using learning Automata and Machine Learning
Classifiers for Network Prediction, studying network dynamics issues and focusing on modeling link
evolution in dynamic network settings.
After constructed a complete link prediction architecture based on learning automata, we put forward a
strategy of optimization to the learning process in the deep reinforcement learning architecture, and
design the full workflow of link prediction.
Consequently, we build the complete learning automata structure and map input data to a category of
data from where the reinforcement learning sent the results to the LA agent which decide if there is
possibilities of the connection to be true or false, there will be link or no link. In this method, there exists
automation learning for individual test link which needs to be predicted.

Published

2020-11-01

Issue

Section

Articles