76 - Increasing In-Class Similarity by Retrofitting Embeddings with Demographics, with Dirk Hovy
NLP Highlights
00:00
The Benefits of Retrofitting for Semantic Classification
It's a very powerful idea, of course. I guess one question that popped in my mind while I read the paper is how expensive is this to do? There's a lot of connections between all pairs of examples of this for the same label. That is an excellent question. And there was something that I was very concerned about at first, because when I first started playing with retrofitting and the original paper makes an explicit point that it is very fast. So now we wanted to apply this to the task of geolocation for tweets and basically saying, all right, each word is connected to all the other words that we've seen in the same place. The training time took between several
Transcript
Play full episode
