These are some codes related to my master thesis. Supplementary documents and codes will be added in future.
More info will come here
This tool converts text based multiple word embedings to binary word2vec file formats. The output file can be used in gensim or other word2vec toolkits. This program has been tested on multiple sense vectors produced by this code and this paper by Neelakantan et al. 2014
The text file can be in two different formats:
<Total Number of words> <Dimensionality> <Number of senses per word> <Does it have the max number of senses per word? (1 or 0)>
<word> <no. of senses>
<global context vector>
<first sense vector>
<second sense vector>
...
For example, the text file can be:
2 3 2 1
bank 3
0.2 0.1 0.3
0.1 0.2 0.2
0.3 0.3 0.3
0.1 0.1 0.1
lemon 1
0.4 0.5 0.1
0.1 0.6 0.2
<Total Number of words> <Dimensionality>
<word> <no. of senses>
<global context vector>
<first sense vector>
<first sense cluster center>
<second sense vector>
<second sense cluster center>
...
For example, the text file can be:
2 3 2 1
bank 3
0.2 0.1 0.3
0.1 0.2 0.2
0.11 0.2 0.21
0.3 0.3 0.3
0.301 0.301 0.299
0.1 0.1 0.1
0.09 0.09 0.098
lemon 1
0.4 0.5 0.1
0.1 0.6 0.2
0.101 0.602 0.201
The output file will have global vector as exact word name. Each word-sense will become "word.." which <n> is an index number. For example if there was 3 sense vectors for the word "bank" in the model, there will be "bank..1", "bank..2" and "bank..3" as 3 sense vectors, and "bank" as the context vector in the output file. If the file was in second format, with supporing cluster centroids, then in output file they will become "bank..1c", "bank..2c" and "bank..3c".