I am working on my first project, I am trying to predict the quality of a software specification requirements. I have 1000 requirements which have been manually labelled on a scale of 1-5 (poor-excellent) quality. My idea is to use doc2vec to in order to predict the quality of a requirement.

I have experimented by following this example: https://towardsdatascience.com/implementing-multi-class-text-classification-with-doc2vec-df7c3812824d

I am not sure if this is a viable solution as the results are not very good, but I'm not sure if this is due to my data or the model.

Each requirement ranges between 1 to 4 sentences, I use taggedDocuments to label the requirements as either 1, 2, 3, 4, 5. Would the correct way to go about this be to label the data according to the 5 classes or should I be labelling them based on each individual requirements.

I have read these answers;

In doc2vec, how to model correctly when many documents share the same label? Doc2Vec - How to label the paragraphs (gensim)

however, I'm still not sure what to do in my particular circumstance so any advice would be greatly appreciated!

Also, I am assuming that word2vec would not be of use to me in this instance as I am interesting in associating similar requirements not the actual words...



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.