Hi, I tried your pretrained consequence in my experiment. I believe this kind of embedding is suitable for dataset with hierarchy structure and can keep the hierarchy information. To evaluate this, I designed a experiment, but got a confusing result.
In the experiment, I choose a small dataset including 1000 words and find the parent words of them in the wordnet. Then, I calculate the distance between the child word and parent word, comparing it to the distance between the child word and other words in wordnet. Surprisingly, I find the distance between the child word and its parent word is relatively large, which I believe should be very small. I count the rank of the distance between the child word and its parent word among all words, and find it is only ranked in the middle.
Hi, I tried your pretrained consequence in my experiment. I believe this kind of embedding is suitable for dataset with hierarchy structure and can keep the hierarchy information. To evaluate this, I designed a experiment, but got a confusing result.
In the experiment, I choose a small dataset including 1000 words and find the parent words of them in the wordnet. Then, I calculate the distance between the child word and parent word, comparing it to the distance between the child word and other words in wordnet. Surprisingly, I find the distance between the child word and its parent word is relatively large, which I believe should be very small. I count the rank of the distance between the child word and its parent word among all words, and find it is only ranked in the middle.
Do you know the reason of the result?
Thanks a lot!