Sociality in Web Forum

Bing Wu

School of Economics and Management, Tongji Univeristy, Shanghai, China

Keywords: Sociality, Web Forum, Exponential Random Graph, Sentiment.

Abstract: We use exponential random graph to examine the generative processes that give rise to widespread sociality

patterns in web forums. We apply the methods to Yahoo finance Wal-Mart message board from 1999 to

2008 to investigate authors’ propensities to establish relationship increase by activity. Research results

shows that although having the lowest percentage in Web forums, medium activity authors are more social

than high activity authors, which shows the consistent pattern with few core members contributing the

majority of content. Considering sentiment, objective authors have the highest sociality, followed by

negative subjective authors, which is proportional to the constitution of author sentiment group. Similar

situation happened in author sociality by class, authors’ tendency to establish relationship is quite different.

We conclude with a discussion of how exponential random graph may contribute to our understanding of

social interaction structure and the processes that create it.

1 INTRODUCTION

A long line of research has shown the importance of

social behavior in web forum, and many of the same

themes continue to be explored today. Despite the

proliferation of web forums, not much is currently

known regarding sociality patterns. Tools are needed

to provide a means of identifying and assessing the

characteristics of social interaction patterns, and how

social interaction patterns further our understanding

of the network-scale sociality in web forums.

Currently Exponential Random Graph Models

(ERGM) provides a better way for construction of a

comprehensive and dynamic model for the following

reasons. Accordingly our research goals are to

develop a general framework for understanding

sociality in web forum based on ERGM and then to

provide a statistically principled approach to the

systematic exploration of several features

simultaneously. And potential applications of this

study can be useful for managing a community and

its members as well (Campbell and Cecez-

Kecmanovic, 2011).

2 LITERATURE REVIEW

To form a comprehensive understanding of our

study problem, a literature review was conducted in

two aspects, including social interaction network in

web forum and corresponding modeling methods.

2.1 Social Network in Web Forum

A web forum details a sequence of discussions

through threads of messages. Therefore, web forums

create implicit ties that connect senders and

receivers in what is often called a “reply network” or

“reply graph.” The reply networks can be analyzed

to identify important relationships (Lee and Lee,

2006). Consequently the social relationships in web

forums can be described as social networks, which

consist of finite sets of authors and the relationships

between them.

As key indicators can be used to understand the

status of social network in web forums, we examine

and classify previous papers in terms of attributes in

web forums, including: author activity, message

sentiment and topics

To establish of any relationships between

measures of web forum discussions and author’s

behavior. The author activity analysis is based on

Activity Theory, which is the idea that human

activity is a dialectic relationship between subject

(person) and object (purpose) (Levy, 2008). In

Activity theory, the constituents of activity are not

fixed, but can dynamically change with conditions

(Kaptelinin and Nardi, 1997). The unit of analysis in

Wu B..

Sociality in Web Forum.

DOI: 10.5220/0004425000540057

In Proceedings of the 15th International Conference on Enterprise Information Systems (ICEIS-2013), pages 54-57

ISBN: 978-989-8565-61-7

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

Activity Theory is an activity directed at an object

which motivates activity.

Sentiment analysis (Das and Chen, 2007);

(Antweiler and Frank, 2004) web forum discussions

is considered to deal with computational evaluation

of expressions of opinion, sentiment, and

subjectivity in text (Pang and Lee, 2008). Broadly

speaking, there are two types of sentiment analysis

approaches. Machine learning-based approaches

require a pre-coded training dataset which consists

of texts and labeled sentiments. Obtaining the

training data and training is a time intensive task

(Aue and Gamon 2005). In contrast, lexicon-based

approaches are faster provided an appropriate

dictionary is available. One of the advantages is that

they can take negation (e.g. not) and intensification

(very) into account. Some well-known lexicons

include Senti Word Net (SWN) (Esuli and

Sebastiani 2006) and the Harvard-IV-4 dictionary

(Tetlock, 2007; 2008).

2.2 Corresponding Modelling Methods

Empirical studies validate a model with real data.

The model is parameterized based on real data, and

the interaction process depicted with estimated

parameters is compared to the real interaction

process. In empirical research of web forums,

findings show that there is a consistent pattern of

participation with a few core members contributing

the majority of content, many peripheral members

contributing infrequently, and a large number of

lurkers (Nonnecke and Preece, 2000) who benefit by

overhearing the conversations of others (Hansen,

2009). The nature the conversation depends largely

on the type of web forum. (Hansen et al., 2010).

Regression analysis evaluates models by testing

their feasibility and equilibrium status based on

mathematical variables and models. Results from

analyses demonstrate that the social context,

including pre-existing social networks, groups, and

intergroup boundaries, significantly constrained the

flow of information interaction pattern across

intercultural CMC (computer mediated

communication, CMC) groups. And in addition the

influence of the social context on CMC

collaboration could be moderated by other

contingent factors such as national culture and

individuals’ expectancies of Internet use (Hichang

Cho, Jae-Shin Lee, 2008)

As social network analysis provides powerful

ways to summarize networks and identify key

people or other objects that occupy strategic

locations and positions within the matrix of links.

The threaded conversation structure in web forums

leads itself well to social network analysis. The basic

properties of social networks include the size,

density, centrality, degree, reach ability (Hanneman,

2001), connectivity (Stocker et al., 2001) and

multiplicity (Emirbayer and Goodwin, 1994).

Pioneering contributions to model social

interaction using ERGM have been made in the

study of sociological implications of friendship

network. ERGM was used to model friendship

formation as a selection process constrained by

individual’s sociality (propensity to make friends),

selective mixing in dyads (friendships within race,

grade, or sex categories are differentially likely

relative to cross-category friendships), and closure in

triads (a friends' friends are more like to become

friends), given local population composition), so that

socio demographic structure and the processes that

creates it can be understood (Goodreau, 2007);

(Goodreau, 2009). Furthermore in order to acquire

sociological implications for single-gender and

cross-gender influences on teenagers’ behavior,

ERGM framework including social network

techniques were used to examine gender clustering

in a complete network of teenagers and their friends

(Kirke, 2009).

3 RESEARCH DESIGN

The Yahoo! Finance Forum is chosen as the test

bed due to the large amount of message postings in

this platform. Wal-Mart was selected due to its

prominence in the market, societal presence, and

active collection of stakeholder groups. As social

interaction pattern in web forum should be

considered over the long-term, the time span for

analysis covers from January 1999 to June 2008.

Each author’s sentiment can be obtained by the

average sentiment of all messages posted or replied

to by whom. In this study, sentiment of each author

will be normalized as three sentiment groups,

negative subjective author (ps1), objective author

(ps2) and positive subjective author (ps3).

The author group depends on the result of key-

phrase extraction for all of his/her all messages

posted or replied to by whom. Thereby the author

class is employees (pt1), investors (pt2) and

customers (pt3) in Yahoo finance forum.

In this study we normalize activity as three

activity groups delegating low activity (pa1),

medium activity (pa2) and high activity (pa3).

SocialityinWebForum

4 RESEARCH RESULTS

Since the dataset spans 10 years, from Jan 1999 to

June 2008, we will use a graphical display, i.e., box

plots. Box plots enclose the interquartile range of the

data in a box displaying the median, giving an idea

about the center, variability, and degree of

asymmetry of a sample. Interquartile range is

between the 75th percentile (upper quartile) and the

25th percentile (lower quartile). Box plots are used

in this study to present the variation for each

measures group with a given degree across the 10-

years span, the vertical solid line in each horizontal

box denotes the median value of each given group

with lower quartile and upper quartile on both ends.

Figure 1 charts the sociality estimates from the

ERGM. As the vertical solid line in each horizontal

box is on the right side of the 0 axis, sociality

estimates for each category are positive.

In the lower part of Figure 1, based on low

activity authors (pa1), the positive median values of

medium activity authors (pa2) and high activity

authors (pa3) mean authors’ propensity to establish

relationship increases by activity. This is consistent

with the pattern of a few core members contributing

the majority of content, similar to research results of

Nonnecke and Preece’s (2000).

The positive median values of negative

subjective authors (ps1) and objective authors (ps2)

denote authors’ preference to post messages

increases based on objectivity (Figure 1). Thus,

objective authors (ps2) have the highest sociality,

followed by negative subjective authors, which is

proportional to the constitution of the author

sentiment group.

The upper part of Figure 1 shows that authors’

tendency to post messages is the strongest among

investor authors (pt2), then employee authors (pt1),

based on the positive median values. This tendency

is consistent with the proportions of the author group

as well.

Figure 1: Sociality estimates plotted across 10 years.

5 CONCLUSIONS

Based upon these findings, we state the conclusions

and implications of this study below. In this case

study, other attributes in addition to author activity,

author sentiment and author class, such as author’s

residence time, reply interval, etc., could be added,

which will be useful for further exploration of

dynamic social interaction pattern in future. In

addition, selective modeling by vertex attributes,

holistic feature modeling and goodness of fit will be

in future ERG modeling, so that complex social

interaction networks in Web forum can be explored

by most appropriate ERG model. Meanwhile an

iterative exploratory technique of progressively

increasing the model complexity should be used in

ERG modeling process. Moreover other experiments

for different test beds are on the agenda of future

direction.

ACKNOWLEDGEMENTS

This work was supported in part by NSFC under

Grant Nos. 71071117.

REFERENCES

Antweiler, W. and Frank, M.: Is All That Talk Just Noise?

The Information Content of Internet Stock Message

Boards. The Journal of Finance, vol. 59, no.3, pp.

1259-295, 2004.

A. Aue and M. Gamon: Customizing sentiment classifiers

to new domains: A case study. Proceedings of Recent

Advances in Natural Language Processing (RANLP),

2005.

B. Nonnecke and J. Preece: Lurker Demographics:

Counting the Silent.CHI Conference, pp. 73-80, 2000.

Das, S. and Chen, M. Yahoo! for Amazon: Sentiment

Extraction from Small Talk on the Web. Management

Science, vol.53, no.9, pp. 1375-1388, 2007.

Deirdre M. Kirke. Gender clustering in friendship

network: some sociological implications.

Methodological Innovations online (4), pp. 23-26,

2009.

Derek Hansen et al.: Analyzing Social Media Networks

with NodeXL. Morgan Kaufmann, 2010.

D. L. Hansen. Overhearing the crowd: an empirical

examination of conversation reuse in a technical

support community. Proceedings of the Fourth

international Conference on Communities and

Technologies. ACM, New York, NY, pp.155-164,

2009.

Emirbayer, M., & Goodwin, J. Network analysis, culture,

and the problem of agency. American Journal of

ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems

Sociology, vol.99, no.6, pp.1411-1454, 1994.

Esuli, A. and Sebastiani, F. SentiWordNet: A Publicly

Available Lexical Resource for Opinion Mining.

Proceedings of Conference on Language Resources

and Evaluation, 2006.

Hannerman, R. Introduction to Social Network Methods,

retrieved from http://www.faculty.ucr.edu/~hanneman/

, 2001.

Hichang Cho, Jae-Shin Lee. Collaborative information

seeking in intercultural computer-mediated

communication groups: testing the influence of social

context using social network analysis. Communication

Research, vol.35, no.4, pp. 548-573, 2008.

John Campbell, Dubravka Cecez-Kecmanovic.

Communicative practices in an online financial forum

during abnormal stock market behavior. Information

& Management, vol.48, no.1, pp.37-52, 2011.

Kaptelinin, V., Nardi, B.A. Activity theory: basic concept

and application, CHI, 1997.

Pang, B. and Lee, L. Opinion Mining and Sentiment

Analysis. Foundations and Trends in Information

Retrieval, vol.2, no.1/2, pp. 1-135, 2008.

Sang Jun Lee, Zoonky Lee. An Experimental Study of

Online Complaint Management in the Online

Feedback Forum. Journal of Organizational

Computing and Electronic Commerce, vol.13, no.1,

pp. 65-85, 2006.

Steven M Goodreau. Advances in exponential random

graph (p *) models applied to a large social network.

Social networks, vol. 29, no.2, pp. 231-248, 2007

Steven M. Goodreau, James A. Kitts, Martina Morris.

Birds of a Feather, Or Friend of a Friend? Using

Exponential Random Graph Models to Investigate

Adolescent Social Networks. Demography, vol.46,

no.1, pp.103-125, 2009.

STOCKER et al. Consensus and cohesion in simulated

social networks. Journal of Artificial Societies and

Social Simulations, http://jasss.soc.surrey.ac.uk/4/4/5.

html, vol. 4, no.4, 2001.

Tetlock P. Giving content to investor sentiment: The role

of media in the stock market, Journal of Finance,

vol.62, no.3, pp. 1139-1168, 2007.

Tetlock P., Teschansky M. and Macskassy S. More than

words: Quantifying language to measure firm’s

fundamentals. Journal of Finance, vol.63, no.3, pp.

1437-1467, 2008.

Xiaojun, W. and Jianguo, X. Exploiting neighborhood

knowledge for single document summarization and

keyphrase extraction. ACM Transactions on

Information Systems, vol.28, no.2, pp.1-34, 2010.

Yair Levy. An empirical development of critical value

factors (CVF) of online learning activities: An

application of activity theory and cognitive value

theory. Computers & Education, vol.51, no.4,

pp.1664–1675, 2008.

Zachary M. Saul_and Vladimir Filkov. Exploring

Biological Network Structure Using Exponential

Random Graph Models. Bioinformatics Advance

Access published July 20, pp. 1-7, 2007.

SocialityinWebForum