![espacio](https://private-user-images.githubusercontent.com/107461889/314465372-fa1f8fc7-e135-4ee9-b54e-b2fb54f135ba.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjM4OTIyNzAsIm5iZiI6MTcyMzg5MTk3MCwicGF0aCI6Ii8xMDc0NjE4ODkvMzE0NDY1MzcyLWZhMWY4ZmM3LWUxMzUtNGVlOS1iNTRlLWIyZmI1NGYxMzViYS5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwODE3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDgxN1QxMDUyNTBaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1iMDlhMmVhOTQ4YTljNGMzNzJlM2M1NzUyNDRmOTE4YzNhZGUwZjkwZTQxYmI1NGRhNmY5ODAzZDYwMjliYjhlJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.aVyi6baZBI2NgRq0hnSp4FZ0ooDXn2S7YS3Ci_iuRNM)
This repository presents a comprehensive evaluation of word embeddings in Spanish with paper and code. We trained word vectors over La Solana using the library fasttext
. You can visualize them following this link to Tensorflow Embeddings Projector. Try searching for 'la_casota', 'ofrecimiento' or 'moje'. Do you have any idea what these words mean in La Solana?
Summary and links for lasolana-embeddings:
Corpus | Size | Algorithm | Vectors | Vec-Dim |
---|---|---|---|---|
Collected corpus | 4.1M | FastText | 29682 | 150 |
Links to the embeddings (#dimensions=150, #vectors=29,682):
- Vector format (.vec) (34.5 MB)
- Binary format (.bin) (34.5 MB)
- TSV format (.tsv) (49.5 MB)
All digitized corpus about La Solana that we have access to:
- Gaceta de La Solana until december 2023
- Noticias de La Solana
- Julián Simón's blog
- García Gallego, José María. El legado Bustillo de La Solana: su fundación, visicitudes, intervención de Costa, asesinato del cura Torrijos, situación actual. (1935)
Corpus Size: over 4 million words Preprocessing: Explained in training_eval notebook.
Implementation: FastText with Skipgram and no sub-words
- min subword-ngram = 0
- max subword-ngram = 0
- neg = 5
- ws = 5
- epochs = 5
- dim = 150
- all other parameters set as default