I have a list of words, these words correspond to labels in news and they are not duplicated. I would like to get a clustering for this list based on topics.

I try with wordnet but I don't know how could get it if I only have a list of unique words and not more information.

Thank you!

  • $\begingroup$ Welcome to DataScienceSE. Normally the way to obtain topic models is to fit a model (typically LDA) on the collection of full texts. This would give you a list of topics together with the distribution of the words by topic (and also the prob of topic given document). Of course this is different from clustering the labels. $\endgroup$
    – Erwan
    Commented Jun 1, 2022 at 17:10
  • $\begingroup$ Thank you @Erwan but my problemn is that I only have a list of words without any context about them, I mean without sentences. LDA model need more information than a list of words? or with this list is fine? $\endgroup$ Commented Jun 2, 2022 at 10:47
  • $\begingroup$ LDA needs quite large documents with some words appearing in multiple documents in order to build good topics, so this can't work in your case, unless you have access to the news articles having these labels. So you're right that you can only measure similarity between words then. I think Wordnet can give you a semantic distance (or similarity) between pairs of words, and then you could use general clustering with a matrix of distances. Another way to measure similarity between words is to use some pretrained word embeddings and measure the similarity (typically cosine) between vectors. $\endgroup$
    – Erwan
    Commented Jun 2, 2022 at 10:59


Your Answer

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

Browse other questions tagged or ask your own question.