0
$\begingroup$

Does anyone know a technique to represent meaning of a given text document, so that two documents having same meaning will have same representations?

Note: Typical systems like vectorization methods are not suitable for this because they only consider presence of vocabulary of the text, not the ordering of words.

$\endgroup$
  • $\begingroup$ You are probably looking for something like semantic embeddings. $\endgroup$ – IonicSolutions Sep 11 '18 at 11:19
  • $\begingroup$ Is there any paper or something published related to semantic embeddings?? can you post something regarding this?? $\endgroup$ – Vajira Prabuddhaka Sep 11 '18 at 11:42
  • $\begingroup$ Latent dirichlet allocation is used to detect topics in documents. You may find it useful. $\endgroup$ – Ian Sep 11 '18 at 12:00
  • $\begingroup$ I suggest you do a first round of Google searching yourself and then ask specific questions. LDA as Ian mentioned is also worth checking out. $\endgroup$ – IonicSolutions Sep 11 '18 at 12:31
0
$\begingroup$

We might need to know what exactly is your goal.

Meaning of a text needs to be a complex structure (whatever approach). So if you want two document with the same meaning to have exactly the same meaning representation, you need to go for a full semantic modelling using some sophisticated approach involving knowledge representation etc.

But how do you measure how much the meaning of two texts is the same / similar?

Typical representationl approach is to have some metric of similarity like cosing distance of two vectors. BTW there is a doc2vec that produces vector representation of the whole text. And doc2vec does care about the context of the words, maybe not strictly about word order but their proximities.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.