Open a1da4 opened 1 year ago
They analyze the relationship between word frequency and word similarity.
Previous works tried to find relationships between frequency and:
This work has a different view.
They use three embedding models:
Two experiments are conducted:
Figure 2 shows that the similarity is high in high-frequency bins.
Figure 5 shows that word vector models encode frequency.
Figure 6 shows that the measured bias is changed drastically by the frequency (especially the frequency of B (he)).
0. Paper
1. What is it?
They analyze the relationship between word frequency and word similarity.
2. What is amazing compared to previous works?
Previous works tried to find relationships between frequency and:
This work has a different view.
3. Where is the key to technologies and techniques?
They use three embedding models:
Two experiments are conducted:
4. How did evaluate it?
4.1 Internal evaluation
Figure 2 shows that the similarity is high in high-frequency bins.
Figure 5 shows that word vector models encode frequency.
4.2 External evaluation
Figure 6 shows that the measured bias is changed drastically by the frequency (especially the frequency of B (he)).
5. Is there a discussion?
6. Which paper should read next?