Information Technology Reference
In-Depth Information
5.5 Problems of Trust Based on Statistical Information
5.5.1 Result of the Experimentation of Trust by the Following
Relationship
We collected the data of 28,830 Twitter users randomly and calculated PersonaRank.
According to the score of PersonaRank, we ranked the users. The top 5 highest
scoring users are listed as follows.
1. “Britney Spears”
2. “ashton kutcher”
3. “Ellen DeGeneres”
4. “Lady Gaga”
5. “Barack Obama”
Almost all persons on this list are famous people. It is natural for many people
follow a famous person. However, these famous people are not always trustworthy
or valuable as a information resource.
The intention to follow someone in the social web and to link some web pages
are different in terms of the trust for those information sources. Most of the
messages on the time line of Twitter contain chatting or jokes. However, some
people appreciate famous people's chat and joke. The value of the information of
the social web and web pages are quite different.
In any case, PersonaRank or other information rating methods based on the
number of followers or friends can not be a suitable index for measuring the trust
for the information sources over social web services.
5.5.2 Experimental Results of Trust Evaluation Based
on the Similarity of Behavior
Next, we confirmed a trust evaluation method which focuses on the similarity
among users. We have focused on the temporal similarity of behavior among
people as an index of similarity. We have selected 578 active users from 28,830
users, who had posted more than 20,000 tweets. We had recorded for 1 week and
counted the number of tweets per hour (Fig.
5.3
).
After correcting that, we calculated the similarity of every pair of the people. The
graph of Fig.
5.4
shows the similarity of the top 30,000 pairs. After checking the
results of our experimentation, almost all the pairs of high similarity turned out to be
a BOT. A BOT is a software robot, which imitates human behavior. We found that
BOTs are prevalent amongst social web services. This result has serious implications
because the interest of this research is concentrated on the pairs of high similarity
and it is very hard to distinguish the BOTs from humans automatically.