Earlier quoted context omitted.
Our problem is not about auto-completion (we're not dealing with that much data to need sophisticated algorithms for that). What we're doing with our NN is ordering the set of results (matches) we already have. In other words, we're assigning a relevance number in [0, 1] to each result, based on the query string and training based on past user choices (clicking a result). In order to maintain some consistency and rob…
Ah ok I think I get it. So herein lies the problem (let me know if any of this is incorrect): you want to encode fragments of a word similarly to completed versions of the word, without doing inference. The easiest way of augmenting your training data into a word2vec embedding with artificially misspelled inputs doesn't seem like it would work considering how many misspellings there are per word. I think the best way…
What we ended up doing for now is a two-dimensional input layer with per-column one-hot encoding of characters (i.e. one character is one column, 128 rows for the ascii alphabet). Then, apply a convolution with kernel dimensions 3x128, which flattens data to one dimesion and combines three neighboring characters. The second part builds an "assiciation" between neighbors, which helps yielding similar outputs for similar word fragments.
This works quite well, except for some nasty limitations:
- Search queries have a hard limit in length, caused by our input layer dimensions
- Due to varying search query length, input nodes on the right side are often unused/zero, leading an weighting bias on the left side when training. That is, the start of search queries receives more attention that the end. But that's not necessarily a bad thing.