Hi dears
If I want to do supervised multi-classes classification of specific domain such as history, using one of the deep learning techniques. Which is the best word embeddings method should I use?
1- Word2vec based on Continuous Bag-of-Words,
2- Word2 vector based on Continuous Skip-Gram Model,
3- Glove.
4- BERT Model
Many thanks in advance.
i recommend you to use all models and to choice model with best metrics. Also you can look at pretrained text encoders on TensorFlow Hub (https://tfhub.dev/google/collections/transformer_encoders_text/1), for examle , BERT.
Instead of naming one though transformers are the best one, you can view some of the comparisons I have performed between Glove, FastText, ELMO here
https://github.com/talhaanwarch/Offensive-Language-Detection
- Glove
- BERT
- I used some of these in one my experimentation here: Conference Paper Adaptation of IDPT System Based on Patient-Authored Text Dat...
I would suggest experimenting with different variations of BERT. I am re-writing my experiments with BERTS and the initial results are very promising.
It depends what kind of task you have (history is very broad topic). There is also 5:
Article Word Replaceability Through Word Vectors
Answer to your query may be very broad, however, I would like to suggest you to follow https://towardsdatascience.com/word-embeddings-in-2020-review-with-code-examples-11eb39a1ee6d
https://aylien.com/blog/overview-word-embeddings-history-word2vec-cbow-glove
- Badges
- Science topic