Detecting Spam on Twitter via Message-Passing Based on Retweet-Relation - Technologies and Applications of Artificial Intelligence - page 63

Fig. 3. Effectiveness of unit time interval selection
4.2 Effectiveness Analysis
Effectiveness of Unit Time Interval Selection. Figure 3 shows the exper-
iment results of applying different unit time interval values to Message-passing
Graph Analyzer. If the unit time interval is too small, the automatically retweet
behavior of spam collusion accounts were not obvious. At the beginning, the
message-passing graphs of some spam tweets were similar to normal tweets.
When the unit time interval is set as five minutes, we get the best precision and
recall rate from our experiment results. If the unit time interval is too large, the
message-passing graphs contained more noisy nodes and effected the graph-based
features. Thus, the precision and recall rate declined slightly.
Fig. 4. The accuracy in different methods