Index of /divisions/hcs/nlplab/resources/swectors

[ICO]NameLast modifiedSizeDescription

[DIR]Parent Directory   -  
[   ]suc.saldo 06-Jun-2016 15:26 398K 
[   ]swectors-300dim.txt.bz216-Jul-2016 04:30 187M 
[   ]create_vectors.py 25-Oct-2016 10:07 1.5K 
[DIR]sentences/ 26-Oct-2016 23:09 -  

Swectors

Swectors

This is the code used for the paper "Towards a Standard Dataset of Swedish Word Vectors", pdf can be found here.

Creating Swedish word vectors

The script takes four parameters, method (cbow or sgns), dimensionality, window size and iterations.

Such as:

python3 create_vectors.py cbow 300 10 40
python3 create_vectors.py sgns 50 10 5

A textfile is created where each the first index of each row is a unique word and the rest of the row is each element of the vector, separated by spaces.

Data

The training set is located in 'sentences', each row corresponds to one sentence. Included is a sample of Göteborgsposten-2013 (100k rows).

Evaluation

Tools and instructions to how the use QVEC-CCA can be found here. For a quick start, simply download the file and add 'suc.saldo', then use the following line to evalute a set a vectors.
./qvec_cca.py --in_vectors /path/to/vecs --in_oracle suc.saldo

Contact

Per Fallgren

Jesper Segeblad

Marco Kuhlmann