Evaluating credibility of interest reflection on Twitter

• Pages: 343-362
• Date: 11 November 2014
• Published date: 11 November 2014
• DOI: https://doi.org/10.1108/IJWIS-04-2014-0019
• Author: Hao Han,Hidekazu Nakawatase,Keizo Oyama
• Subject Matter: Information & knowledge management,Information & communications technology

Evaluating credibility of interest

reection on Twitter

Hao Han

Department of Information Sciences, Kanagawa University, Hiratsuka,

Japan, and

Hidekazu Nakawatase and Keizo Oyama

Digital Content and Media Sciences Research Division,

National Institute of Informatics, Chiyoda, Japan

Abstract

Purpose – The purpose of this article was to conrm whether users’ interests are reected by tweeted

Web pages, and to evaluate the credibility of interest reection of tweeted Web pages.

Design/methodology/approach – Interest reection of Twitter is investigated based on the context

of sharing behavior. A context-oriented approach is proposed to evaluate the interest reection of

tweeted Web pages based on machine learning. Some different distribution models of similarity are

present, and infer whether tweeted Web pages reect respective users’ interests by analyzing user

access proles.

Findings – The analysis of browsing behaviors nds that many users partially hide their own

concerns, hobbies and interests, and emphasize the concerns about social phenomenon. The extensive

experimental results showed the context-oriented approach is effective on real net view data.

Originality/value – As the rst-of-its-kind study on evaluating the credibility of interest reection on

Twitter, extensive experiments have been conducted on the data sets containing real net view data. For

higher accuracy and less subjectivity, various features are generated from user’s Web view and Twitter

submission background with some different context factors.

Keywords Twitter, Credibility, Browsing behavior, Integrated analysis, Net view data,

Social network

Paper type Research paper

1. Introduction

Twitter (www.twitter.com/) is the most popular micro-blogging service and plays an

important role in social network. The service has rapidly gained worldwide popularity,

with over 500 million registered users (Twitter Turns Six, 2012) and generating over 400

million tweets daily (Celebrating Twitter7, 2013). It attracts more and more users to

share the interesting Web pages such as news (Kwak et al., 2010). The Web pages are

effectively spread and shared as a form of shortened URLs (http://t.co/) in the posted

messages (tweet) by reproduction (retweet and reply) through the users subscription

(follower).

We gratefully acknowledge the data provided by Yamana Laboratory (Department of Computer

Science, Waseda University, Japan). We also acknowledge the technical support provided by Feng

Xiao (Department of Computer Science, Tokyo Institute of Technology, Japan). This work was

supported by a Grant-in-Aid for Scientic Research A (No.22240007) from the Japan Society for

the Promotion of Science (JSPS).

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/1744-0084.htm

Evaluating

credibility of

interest reection

343

International Journal of Web

Information Systems

Vol. 10 No. 4, 2014

pp. 343-362

© Emerald Group Publishing Limited

1744-0084

DOI 10.1108/IJWIS-04-2014-0019

The rapid development of Twitter stimulates social media-oriented researches, and

personalized community recommendation such as Web page recommendation becomes

a popular issue and a hot research topic. Personalized recommendation is based on the

analysis of users’ concerns and interests. There is a comprehensible but unproven

assumption:

Users access interested webpages that attract their curiosities and attentions, and share them

on Twitter. If a user shared ten sports webpages and one politics webpage, this user should

have more interest in sports than politics. So, it is natural to nd out users’ interests and

concerned topics by analyzing their shared webpages.

This assumption is widely reected in many recommendation approaches as common

sense.

However, an important difference between Twitter and other services, especially

traditional blogs, is ignored. As a quick and convenient way of sharing Web pages, most

tweets for sharing Web pages contain few or even no original opinions and comments of

the users. They copy and paste URLs into tweets or just click embedded Tweet Buttons

to share the Web pages. A tweet can be posted within an extremely short period of time

such as 5 seconds, which is far inadequate for posting a traditional blog article. This

quick pace of Web life brings many uncertain elements in sharing behaviors. For

example, some users may share Web pages just for attracting more retweets, replies and

followers on Twitter (Halpern) without seriously considering whether these Web pages

bring concerns or interests to themselves since sharing costs few seconds and brings

nothing harmful to their Web proles. Thus, “big data has big noise”, as users may tend

to post many “not worth reading” tweets (Andre et al., 2012), we cannot fully believe that

the tweeted Web pages reect the users’ interests.

In this paper, we present an in-depth analysis of the user browsing behaviors on

Twitter by using net view data, Twitter data, and Web pages as data sets. Our analysis

emphasis is laid on Web pages provided by external Web sites instead of retweets, posted

photos or chatters. We focus on two aspects:

(1) conrming whether users’ interests are reected by tweeted Web pages; and

(2) evaluating the credibility of interest reection of tweeted Web pages.

Regarding point (1), we investigate the context of sharing behavior and explain the

investigation results. As for point (2), we propose a novel approach to judge the

credibility of interest reection of tweeted Web pages based on clustering and support

vector machines (SVMs). Extensive experiments have been performed on real net view

data to show our approach is exible and effective.

The contributions of this paper are summarized as follows:

• Thedifferent sources of data are integrated and the analysis range is extended to

the general accessed Web pages (not limited to tweeted Web pages). To realize the

effective and comprehensive comparison of the difference between the accessed

Web pages and shared Web pages for respective users, we present some different

distribution models of similarity between those Web pages during a short time

period (e.g. 1 hour), and infer whether tweeted Web pages reect respective users’

interests by analyzing user access proles (all accessed Web pages) during a long

time period (e.g. one month).

IJWIS

10,4

344