Peer Effect in Online Communities for Modeling User Behavior and Group Dynamic Evolution Research based on Deep Reinforcement Learning Algorithms

Abstract

An increasing number of people are joining online technology communities to meet their personalized learning needs with the development of online communities. In this manuscript, a Graph Wavelet Gated Recurrent Neural Network (GWGRNN) optimized with the Humboldt Squid Optimization (HSO) for predicting the user behavior in online communities with group dynamic evolution (UBOC-GWGRNN-HSO) is proposed. Initially, the data is collected from User Behavior on Instagram (UBI) datasets. Then, the data is fed to a Multi-window Savitzky-Golay Filter (MWSGF) based preprocessing process. Then, the preprocessed data’s are fed to GWGRNN to classify the user behavior in online communities like, positive, negative and neutral behavior. The weight parameters of GWGRNN are optimized using HSO. The proposed UBOC-GWGRNN-HSO is implemented in python, effectiveness assessed by several performance metrics, like accuracy, error rate, F1-score, precision, recall, sensitivity. The proposed method error rate attains 3% positive, 4% negative and 2% neutral of the user behavior in online communities. The proposed method shows better results in all existing systems like Sarcastic user conduct from social networks using firebug swarm optimization-based long short-term memory (SUBSM-FSO-LSTM), Particle Swarm Optimization for Detecting Social Spam Bots Users in Twitter Network (DSUTN-PSO) and predicting customers purchase behavior using deep neural network (PCPB-DNN). From the result it is concludes that the proposed UBOC-GWGRNN-HSO method based error rate lower than the existing methods.