Questions Emoji

1. What is the purpose of the DeepMoji model? Can you see other uses for this model (not mentioned in the paper)?

2. Is the model capable of working with texts longer than 70 tokens? If yes, what challenges does it bring (if any)?

3. Does the model (described in Section 3.2) use self-attention (as described in the Transformer paper)? What are the similarities and differences?

4. Table 3 shows Top 1 and Top 5 accuracy of four models in a "64-emoji labeling" task. Guess what would be the Top 1 and Top 5 accuracy of the Random and DeepMoji (d=1024) models when instead of 64, we would use only the 8 emojis listed in Figure 5 (in Appendix A.3).

5. Can we say DeepMoji has a super-human accuracy in predicting the polarity of a tweet (positive vs. negative)? Why?