I'm currently working on the task of measuring semantic proximity between sentences. I use fasttext train _unsiupervised (skipgram) for this. I extract the sentence embeddings and then measure the cosine similarity between them. however, I ran into the following problem: cosine similarity between embeddings of these sentences:

"Create a documentation of product A"; "he is creating a documentation of product B"

is very high (>0.9). obviously it because both of them is about creating a documentation. but however the first sentence is about product A and second is about product B and I would like my model to understand that and emphasise on those product names since they are key words in sentences. Which type of model would be more suitable for my case? Is BERT for example better for it and why?

  • $\begingroup$ this is an open question. And as far as theoretical results are concerned (eg NFL theorems) it is unlikely one method fits all $\endgroup$
    – Nikos M.
    May 26, 2022 at 15:04
  • $\begingroup$ Huggingface has hundreds of models to detect sentece similarity. You could try the top 3 of them and see which one fits best to your data. huggingface.co/… $\endgroup$ May 27, 2022 at 13:42


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.