76 - Increasing In-Class Similarity by Retrofitting Embeddings with Demographics, with Dirk Hovy
NLP Highlights
00:00
How to Predict Author Age and Author Gender From Text
The idea behind this is something called homophily, which is also sometimes translated as birds of a feather flocked together. What we're doing here in general is trying to take the word embeddings that we've induced on the training data and then increase the similarity within each of the classes that we have. So for example, for age prediction, we have 10 classes and we're trying to make people within each class speakers within each class more similar to each other.
Transcript
Play full episode
