Please use this identifier to cite or link to this item:
Title: Opinion influence modeling in social media
Authors: Chen, Chengyao
Degree: Ph.D.
Issue Date: 2018
Abstract: Online social media have gained a lot of popularity and experienced a fast growth in the past decade. The emergence of social media offers ordinary persons remarkable opportunities to create messages expressing their opinions. Besides, by establishing relationships with others, people can easily convey their opinions to others. Opinion influence produced by social interaction becomes an important factor for people to adapt their behaviors and make decisions. Understanding opinion influence would greatly benefit a variety of marketing activities, such as spread of ideas, public opinion monitoring, and intervention. This thesis aims to provide insights into opinion influence modeling from three components: user interaction, temporal dynamics and the exchanged content. Twitter datasets, which contain the users' opinion traces and user network structures, are collected for the study of opinion influence. The first work investigates the temporal properties of opinion behaviors. Opinion influence is produced by long-term interactions, where a user continuously collects opinions from neighbors and further changes her/his own opinions accordingly. The temporal dynamics is of great importance for uncovering the underlying mechanism of opinion influence, but is ignored in the current research work. Here, we propose a temporal opinion influence model, which is able to track the opinion dynamics of each individual user and uncover opinion influence through correlating opinion dynamics of connected users. Specifically, we propose two indicators to capture the effects of social interaction on the opinion formation, including friend effect and opinion effect. The second work delves into the textual content exchanged during communication. The textual message, as the medium of social interaction, provides the foundation to understand communications between users. Rooted in neural network technology, a content-based opinion influence model is proposed to study how opinion influence is driven by the content. Apart from the exchanged content, we also consider the identities of users involved in communication. Each user is characterized by a personal identity and a social identity. A joint learning framework is developed to detect the social identities of users and models opinion influence concerning different user identities at the same time. This work first goes a step further to introduce the content into opinion influence modeling. Its novel idea of integrating personal images in the understanding of user opinion influence also contributes. In the third work, all the components explored in the above two studies, including the temporal dynamics of user interactions and the content included in the exchanged texts, are carefully considered. Inspired by the advances of the recurrent neural network in sequence modeling, a sequential content-based opinion influence model is developed. It offers to predict opinion words other than opinion sentiment, which may benefit marketing analysis in a more comprehensive manner. This work provides a complete and effective understanding of the opinion influence process. It can be further extended to model the content-based user dynamics in other scenarios. We conduct a systematical study to understand opinion influence on social media from three components. Our study benefits a variety of marketing activities, including advertisement dissemination, optimization of product impact and other business intelligence related applications. Besides, other complex dynamics of human behaviors, from buying behaviors in the business to voting behaviors in the politics can be unrevealed by continued extension of the proposed framework.
Subjects: Hong Kong Polytechnic University -- Dissertations
Social media -- Social aspects
Social interaction
Influence (Psychology)
Pages: xx, 142 pages : color illustrations
Appears in Collections:Thesis

Show full item record

Page views

Last Week
Last month
Citations as of May 28, 2023

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.