Information Technology Reference
In-Depth Information
We use the other two methods to evaluate the influence of node: UserRank [9] and
TURank [15].These methods are based on PageRank algorithm. The core idea of
UserRank is that the numbers of friends is an important index of influence. The
TURank could to reveal the relationship between user and information and obtain a
more accurate result of ranking. The experimental results are as follows.
Fig. 5.
Comparision of three methods
We selected the top ten users as shown in TABLEII. The value of IR gradually
decreased from node 1 to node 10. Observed from the curve of UserRank, the first four
nodes' influence is decreasing, but the influence of subsequent node is increasing, the
influence of node 9 is very large. Why trends of the two curve is not consistent? The
most important reason is that the two methods considering the problem from different
angles. The method of UserRank focuses more on the number of friends. For example,
the experiment shows that node 9 has 876 friends so that its influence is larger.
Because of the existence of zombie fans, the node has a lot of friends and its influence
unnecessarily large. Compared with IR model, UserRank is not considered
comprehensively, the result of UserRank is not very accurate.
Observed from the curve of TURank, the first four nodes' influence is decreasing,
the influence of subsequent node vary irregularly. The trends of the IR model and
TURank are not consistent. The method of TURank pays more attention to interactive
behavior and information of user own. As is shown in the Figure5, the node 6 has a
large value of TURank; because of the node has more interactive behavior. The
method of TURank ignored the number of fans and its initial value should be set
artificially. Therefore, the results of two methods are not completely consistent.
Compared with TURank, the result of IR model meets the actual better.
From the results of these experiment can be seen, the evaluation of nodes influence
based on user's attribute and behavior considering many factors, and the result is more
realistic. This method has certain applicability.
6
Conclusion
This paper proposes the model of IR; it can be comprehensive consideration attributes
and interactive behavior. Firstly, we calculate the node attribute value AR, and then
