Model: Word2Vec
Lang: IT
Model description
This model is a lightweight and uncased version of Word2Vec [1] for the italian language. It's implemented in Gensim and it provides embeddings for 560.509 uncased italian words in a 100-dimensional vector space, resulting in a total model size of about 245 MB.
Training procedure
The model was trained on the italian split of the Wikipedia dataset (about 3.7GB, lowercased and pre-processed) for 10 epochs, using a window size of 5 and including words with a minimum count of 10, with an initial learning rate of 2.5e-3
Quick usage
Download the files in a local folder called "word2vec-light-uncased-it", then run:
from gensim.models import KeyedVectors
model = KeyedVectors.load("./word2vec-light-uncased-it/word2vec.wordvectors", mmap='r')
model.most_similar("poesia", topn=5)
Expected output:
[('letteratura', 0.8193784356117249),
('poetica', 0.8115736246109009),
('narrativa', 0.7729100584983826),
('drammaturgia', 0.7576397061347961),
('prosa', 0.7552034854888916)]
Limitations
This lightweight model is trained on Wikipedia, so it's particularly suitable for natively digital text from the world wide web, written in a correct and fluent form (like wikis, web pages, news, etc.).
However, it may show limitations when it comes to chaotic text, containing errors and slang expressions (like social media posts) or when it comes to domain-specific text (like medical, financial or legal content).
References
[1] https://arxiv.org/abs/1301.3781
License
The model is released under Apache-2.0 license