For a recent research paper, I plan to perform the following, for which I'd kindly ask for your advice.

I obtained a set of a few thousand job ads. I now want to analyse how and whether these job ads include 'content' that has been previously specified in another research paper as individual 'categories'. To make things more precise, there are about 15 existing categories, each of which contains descriptions that explain the category in 2-4 sentences.

Now, I want to understand, which and how many job ads cover the aspects described in each of the 15 categories. A result could be, for example, job ad #1 contains content that matches (or comes close) to the descriptions of categories 2, 5, 8 but misses content that would allow any reference to the remaining categories.

In case you got any references or advice how to approach this task, please let me know. I would suspect that the best approach would be a supervised learning approach.




In terms of data understanding, I would recommend an unsupervised approach first. For example you could ...

  • .. perform tf/idf vectoring and build a simple tag cloud
  • .. perform Latent Dirichlet Allocation to get an overview of latent topics in your data (like clustering and interpret the categories)
  • .. train a word embedding (e.g. word2vec) on top of your data. Then perform a dimension reduction (PCA) to visually explore your data. (You could find semantically and syntactically clusters)

After the unsupervides data exploration you could perform a text classification based on classical approaches (e.g. bag of word models) or with neural nets (=supervised).


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.