使用NTLK与主题相关的单词（另一个单词）

Question

I'm working on building my own virtual assistant based on intent<->action mechanism and want to run some NLP on sentences that my users write.

I want to find out words related to a topic(another word) so that I can define an intent, for example:

If a user asks: Will it rain tomorrow? What's the weather today? is it sunny? Is if going to be windy this afternoon?

I want to be able to say that rain, weather, sunny, sun are related to the intent called weather so that I can communicate with the relevant API and retrieve the requested information.

I'm currently working with Python 3 and NLTK but using synonyms, path to parent and similarity doesn't really do the trick:

wordFromList1 = wn.synsets('weather')[0]
wordFromList2 = wn.synsets('cold')[0]
value = wn.wup_similarity(wordFromList1, wordFromList2)
print(value)
---------------------------
0.1

You can see that the similarity here for cold and weather is really weak. Any suggestions?

Thanks,

Answer 1

Yeah, don't use wordnet for this! it might be (very loosely) correlated to solving what you want but will yield horrendous accuracy and recall unless you engineer the hell out of it. For practical results what you need to do is look at things like LSA, LDA or word embeddings, something along those lines