Skip to main content
. 2021 Jul 13;15:659410. doi: 10.3389/fnhum.2021.659410

Table 9.

Relation detection results of the multi-modal model using the convolutional EEG decoding component.

Randomly initialized GloVe BERT
Model P R F1 (std) P R F1 (std) P R F1 (std)
Baseline 0.404 0.525 0.452 (0.04) 0.501 0.609 0.539 (0.05) 0.522 0.788 0.623 (0.05)
+ noise 0.424 0.299 0.342 (0.06) 0.547 0.441 0.486 (0.06) 0.532 0.493 0.511 (0.07)
+ ET 0.415 0.307 0.345 (0.08) 0.447 0.413. 0.428 (0.07) 0.558 0.665 0.593 (0.13)
+ EEG full 0.225 0.225 0.225 (0.06) 0.548 0.408 0.464 (0.07) 0.647 0.664 0.650 (0.09)
+ EEG θ 0.437 0.380 0.400 (0.05) 0.620 0.493 0.547 (0.05) 0.721 0.698 0.707+ (0.03)
+ EEG α 0.372 0.366 0.352 (0.12) 0.509 0.433 0.461 (0.12) 0.661 0.697 0.675+ (0.08)
+ EEG β 0.394 0.328 0.338 (0.09) 0.627 0.479 0.541 (0.05) 0.643 0.646 0.640 (0.11)
+ EEG γ 0.405 0.363 0.366 (0.09) 0.646 0.490 0.555 (0.04) 0.667 0.699 0.679+ (0.06)
+θ+α+β+γ 0.324 0.227 0.257 (0.11) 0.460 0.436 0.437 (0.14) 0.610 0.562 0.584 (0.05)

We report precision (P), recall (R), F1-score and the standard deviation (std) between five runs. The best results per column are marked in bold. Significance is indicated on the F1-score with asterisks: * denotes p < 0.05 (uncorrected), + denotes p < 0.003 (Bonferroni corrected p-value).