The Design And Implementation Of Tweet Popularity Estimation Based On Deep Learning

Nowadays,social networks are in rapid development,and the number of users has reached an unprecedented scale.Due to the promotion and popularization of social networks,our lives are closed connected.In social media,users can not only post text,audio and video in various forms,but also share content posted by others and spread them through the network connected by user relationships.Tweet popularity,as a quantitative indicator to measure the degree of people's attention for the given tweet,can be defined as the number of clicks thumbs of the video,the number of participants of the topic and so on.As an important issue in social network,tweet popularity estimation can help organization strengthen advertisement push and public opinion propagation in a short time,and individuals broaden their appeal and acquire more reputation.This thesis proposes a tweet popularity estimation algorithm based on history tweet information,which can rely on the evolution rule of historical information and combine the characteristics of the current tweet to accurately predict the popularity of the current tweet.The model consists of four parts:First,the model establishes a deep learning model for tweet popularity prediction based on the temporal characteristics of history information;Second,the model designs a content attention mechanism aiming at the impact bias of historical tweets on the current tweet.Third,the model designs a time aware mechanism for the incompatibility of deep model and time interval;Finally,the proposed model employs Poisson regression model to obtain the overall loss for tweet popularity estimation by analyzing the metric characteristics of the tweet popularity.In extensive experiments,this thesis compares the proposed algorithm with other baselines and the state-of-arts.Experimental results show that the proposed algorithm is superior to other algorithms in the evaluation metric of MAE and SRC.Thus,it also proved the rationality and accuracy of the proposed algorithm in this thesis.In addition,this thesis also discusses the impact of different hyperparameters to the proposed algorithm,in order to adjust the algorithm to optimal performance.