The test set contains 300 suspected rumor microblogs and 500 related comments for
each suspected rumor microblog.
Spam comment generally has a low similarity with the original microblog in
content and is often a declarative sentence, with a high proportion of noun. Therefore, it
can be filtered according to the characteristics of similarity and noun proportion.
4.2 Experimental Result
When the rumor detection threshold is 0.3, the result of rumor detection is shown in
Table 2.
The experimental results of the evaluation criteria are shown in Table 3.
Fig. 2. CNN-LSTM rumor detection model fusing sentimental polarity
152
S. Lv et al.