The challenge I have is a bunch of questions(long text) that are closely matching with an user utterance (short text). I have tried cosine similarity & Tf-iDF, BM25,Jaccard similarity, etc., but none of these give desired results.

For instance, my question list is a) how do I transfer from XYZ account b) will I be able to transfer from XYZ account with zero fees c) how do I receive some money from XYZ account

utterance - "transfer XYZ account" should result in close similarities between all these.

However an utterance - "transfer XYZ account with some fees" should result in b alone (with other two questions having very less similarity scores)

Another challenge that I face is ordering of these entities. From A --> B is treated similarly as B --> A.

  • $\begingroup$ how about word2vec or bert? $\endgroup$ – Nikos M. May 6 at 15:49
  • $\begingroup$ Thanks for suggesting Nikos - Word2Vec is giving me false positives while BERT gave false negatives. $\endgroup$ – Bharadwaj May 7 at 0:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.