How do I suggest tags to the user based only on the title of a list?

Question

The Problem:

I need to suggest tags to the user based only on the title (5-15 words) of a list they are about to create.

We have around 30 pre-determined tags -

Gaming, Movies, TV shows, Documentaries, Books, Music, Art, History, People, Adventure, Sports, Cooking, Travel, Places, Food, Drinks, Fitness, DIY, Technology, Science, Cars, Bikes, Comedy, Shopping, Clothes, Fashion, Photography, Nature, etc.

So, for example, for a list with title 'Most expensive fine-dine restaurants around the world' suggested tags could be (Food, Places, Drinks, Travel) .

It does not need to be super accurate, just needs to work satisfactorily well, I am sure it would get better as more training data comes in from our users. I don't have any training data for supervised learning yet.

I find myself lost in the vast space of Machine Learning and Natural Language Processing. It would be very helpful if someone can suggest what methods/algorithms/libraries I should use for this specific task, and the background reading I should do before it.

Thanks

Answer 1

You can use word2vec. Get some pretrained model, calculate vectors for tags. Then calculate vector for new title. Find cosine similarity between title vector and each tag vectors. Take for description tags which similarity to title greater then some threshold.