Baseline performance

[chijames]

Hi,

Have you checked the result of predicting no topics at all? (the model predictions are all 0s). I might have done something wrong on my end about data preprocessing, but the results seem to be better than the proposed method on the ICSI dataset.

Thanks.

[chijames]

Again, this might be my mistake, but the segmentations seem to occur only at the first 20% or so of the whole meeting. Did you also observe this behavior? Thanks.