Word Embeddings for testing:

https://onedrive.live.com/redir?resid=4696410962B7B2D9!4108&authkey=!AGh3D_9fneXGYnI&ithint=file%2cgz

Please use tool provided here to generate sense vectors for reproducting the RETRO model:

https://github.com/sjauhar/SenseRetrofit

All codes in my Essay could be found at:

scr/sense_fusion/ folder.

The models correspondent files are:

CBOW: simple.py

Sense Pooling (SPM): simple_linear.py

Optimized Sense Pooling (OSPM): opt1_linear_new.py

Neural Network Model (NNOSPM): NN_hidden.py

Word Vectors used in this model could be download at:

The Result of NN model with C(s_i)^{(t-1)} - C_(s_i)^{(t)} > epsilon (= 1e-8):

DEVELOPMENT NOTES

H28 May 28th,: Project reopen for my essay on NAIST.

there are 48 different words in R&G-65