Social Network Sentiment And Behavior Analysis Based On Representation Learning

As a result of their massive,diverse,and increasing user bases,social network platforms act not only as a way for interactions and communications among people but also as a way for people to share feelings and express attitudes.These large amount of up-to-date information have emerged as an important source for opinion mining,sentiment analysis and behavior behavior.The links between users constitute the topology of the social network;the content published by users is the social network information,and the interaction between users make the information disseminate in the social network.The research and application of the above mentioned large scale social network has become a hot research field in computational social science and artificial intelligence in recent years.The sentiment and behavior analysis in social networks is of great significance for many tasks,such as stock market prediction,recommendation systems,and inference of public mood about social events.In recent years,embedding representation learning,which uses neural network models to represent nodes in the network as a set of low dimensional,continuous,dense vectors,has been widely used in the field of social network analysis.This kind of representation method is a new way for the research and development of social computing tasks.In this thesis,we focus on the study of sentiment analysis and behavior analysis based on representation learning.The main contents of our works include the following four aspects:(1)Different from explicit sentiment words indicating sentiment polarity,implicit sentiment analysis is a more challenging problem due to the lack of sentiment words,which makes it inadequate to use traditional sentiment analysis method to judge the polarity of implicit sentiment.In this paper,we propose multiplex network embedding for implicit sentiment analysis(MEISP)model.In this model,we make use of a heterogeneous network to model the text and learn representations of users,entities they commented on and words they used simultaneously.In particular,by performing the word graph based text level information embedding and heterogeneous social network information embedding(i.e.user social relationship network embedding,and user-entity sentiment network embedding),the proposed scheme learns the highly nonlinear representations of network nodes,explores early fusion method to combine the strength of these two types of embedding model,optimizes all parameters simultaneously and creates enhanced context representations,leading to better capture of implicit sentiment polarity.(2)More people are used to express their attitudes on different entities in social networks,forming user to entity sentiment links.These sentiment links imply positive or negative semantics.Most of current user sentiment analysis literature focuses on making a positive,neutral,or negative sentiment decision according to users’ text descriptions.Such approach,however,often fails to retrieve users’ hidden real attitudes.We design a powerful sentiment link analysis framework named graph network embedding for sentiment analysis(NESA).It first utilizes deep autoencoder to learn joint representations of users’ social relationship by preserving both the structural proximity and attribute proximity.Then,a multi view correlation learning based autoencoder is proposed to fuse the joint representation and the user-entity sentiment polarity network.By jointly optimizing the two components in a holistic learning framework,the embedding of network node information and multi-network contents is integrated and mutually reinforced.(3)User behavior prediction with low dimensional vectors generated by user network embedding models has been verified to be efficient and reliable in real applications.However,existing graph representation learning methods mainly focus on homogeneous and static graphs and cannot well represent the real world social networks that are heterogeneous and keep evolving.To address this challenge,we propose a dynamic heterogeneous user behavior analysis network(DHBN)model,which applies graph network embedding to fuse multi-networks information by considering their heterogeneity and evolutionary patterns over dynamic networks.In particular,by separately performing user social relationship embedding,node attribute embedding and user behavior embedding,the proposed scheme learns the highly nonlinear representations of network nodes;and then we explore recurrent neural networks based on attention mechanism to capture the networks’ dynamic evolution.(4)Social network structure and text sentiment play an important role in understanding the information dissemination mechanism of social network.This thesis analyzes the influence of sentiment intensity and social relationship on users’ retweet behavior in social network.The proposed graph representation learning for user retweeting behavior prediction(RBPNE)model is based on the graph convolutional network,which considers the users’ social relationship strength,the sentiment strength of the dissemination content in the update process of the embedding representation,and the distance between different user node and the target mesage node,at the same time,characteristics of message propagation from top to bottom along the user relationship chain are extracted to explore and find out which message a user will retweet.To sum up,by extracting a variety of graph relationships and designing a multi class heterogeneous relationship extraction algorithm,this thesis studies the embedding representation model of heterogeneous data,establishes a representation learning method for heterogeneous dynamic environment,and forms an overall analysis method of users’ sentiment and behavior in social networks,and further solves the problem of social network analysis in time domain and relationship domain.