Many research papers use the word "encoding" in many ways. One way is that they use the word "encoding" for models that do convert the text into a vector (say text encoding or sentence encoding).

But, afaik, "word embedding" is the formal one to be used in case of conversion of text to vector.

I am thinking like this: If there is text of $n$ words then the text can be represented by a large vector, which is formed from $n$ word embedding vectors. Thus, word embedding is the only useful technique in converting text to vector.

Is there anything like "sentence embedding" or "text embedding" or "text encoding" in contrast to what I think (i.e., does not use "word embedding" in converting to vector)?

If no, is there any reason to use the word " (text) encoding" instead of "(text) embedding"?

For example: I want to find a vector that represents the sentence "She is studying". Then I need to find word embeddings for the three words.


$$she: [1, 4, 6, 9],$$ $$is: [5, 7, 2, 3],$$ $$studying: [89, 54, 12 , 3]$$ represent word and its corresponding embedding, then I am thinking that the sentence encoding gives $$"\text{She is studying}": [1, 4, 6, 9, 5, 7, 2, 3, 89, 54, 12 , 3]$$

Or is "text encoding" can be a different (to what I think) method that is independent of word embeddings?


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.