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).
SocialityinWebForum
55