The way for utilizing neural network as transformation from string to vector.
The distance between two vectors is similar to Levenshtein. This method can be used to draw words on a plot.
I used recurent neural network. It lets process letters one by one.
The traning process:
- Take two random words
- Calc Levenshtein distance between words (L)
- Calc distance between vectors after transformation (D)
- Lets take such position that minimize (L - D)^2
- Start back propogetion for that position