Trust in Online Collaborative IS - Community-Built Databases: Research and Development - page 356

users. They categorize users into four groups - administrators, anonymous users,
registered users, and blocked users - and assign a static reputation value to each
group. In [ 37 ], authors consider dynamic and individualized reputation values for
registered users, but assign a static and nonindividualized reputation value to anony-
mous users.
14.3.2.4 Resistance to Attacks
According to the proposed models, users increase their reputation when their con-
tributions to the wiki pages survive. The robustness of the models are highly
dependent on when the reputation gain events are triggered. Assume that the reputa-
tion of a user increases immediately after he/she inserts some content; if the page is
revised only after a long period of time, the user will have an increased reputation
throughout the period, even if his/her contribution is of poor quality. One solution to
this problem is to postpone the reputation increase until the contribution is reviewed
by another user. Although this solution solves the previous problem, the reputation
model becomes vulnerable to a Sybil attack, 8 whereby an attacker has multiple
identities and can follow up his/her own edits. To overcome both problems at
once, we postpone the reputation increase until a high reputation user (e.g., admin)
approves the corresponding page. Therefore, in the proposed models, a reputation
gain can be triggered only when an admin submits a new revision. One may argue
that this reliance on the limited number of admins as outside authorities might reduce
the accuracy or scope of applicability of the proposed models. However, as shown in
Table 14.6 , in Wikipedia we have large number of good users who contribute
actively to Wikipedia pages. Thus, enlarging the pool of authorities beyond admins
to include these good users to validate the quality of insertions may provide an
efficient solution, especially for pages with high edit rates.
Among related work, Chatterjee et al. [ 52 ] have addressed the attack resistance
problem by extending their previously presented model [ 37 ]. Although the
extended model is resistant to the aforementioned attacks, it is considerably more
complex than the original model. Since we do not consider the stability of deletes
and reverts and we ignore the side effects of reverts, our proposed models are not
prone to other kinds of attacks, such as delete-restore or fake followers [ 52 ].
Another issue in the proposed models is that reputation gains happen without
giving any consideration to the quality of the page that a user contributes to. In [ 53 ],
the authors make two assumptions: (1) the quality of a wiki page depends on the
reputation of its contributors; and (2) the reputation of a user depends on the quality
of the pages he/she contributes to. Although the first assumption is often true, the
second assumption is more debatable; furthermore, it also increases the vulnerabil-
ity of the model against some attacks. Our study of Wikipedia shows that vandals
are more active in high-quality pages. For example, the average RDR associated
8 http://en.wikipedia.org/wiki/Sybil_attack