Semantic Matching Query to Wiki

[KChalk]

Queries should be answered via the most strongly related documents.

1. Currently:
   1. distance calculation between query and documents is implemented via ngrams
   2. learned embeddings representations of documents exist but are not used
2. Next Steps:
   1. represent queries via learned embeddings (done)
   2. implement comparison of q and doc learned embeddings (brute force done)
   3. reduce computational expense of searching document representations
3. Stretch:
   1. more complex document representations
   2. return n best docs (done) and search all for best answer
   3. improve models choices and embedding processes (move away from averaging)

[KChalk]

2.iii: agglomerative clustering could make for n-ary search.