Table 3 Results Don't Match The Discussion

[ShichangZh]

Hello,

I am reading your BGNN paper and get a bit confused by the results in Table 3 and the discussion in the paragraph following the table.

The sentence on top of page 7 says that

"The other two datasets, Slap and DBLP, have sparse bag-of-words features that are particularly challenging for the GNN model. On these two datasets, GBDT is the strongest baseline. Moreover, since FCNN outperforms GNN, we conclude that graph structure does not help"

However, seems like the numbers of FCNN (Slap: 0.759, DBLP: 0.623) in Table 3 are not better than GNNs (Slap: > 0.8, DBLP: > 79 except GCN)? May I ask why? Am I missing anything?

I appreciate your help and thank you for the hard work for this interesting paper.