https://datascience.stackexchange.com/a/7000
https://www.kaggle.com/c/facebook-recruiting-iii-keyword-ext...
1. Use nltk to get all the nouns etc. from the topic.
2. You can use LDA/TF-IDF (gensim) for your questions to extract the most common topic.
3. Use cosine similarity once you have the corpus built to tag a question to the right category.
Look into gensim for more details for the topic modeling.