What's the best way to identify keywords?

Question

natural language processing

What is the best way to define message keywords in a message flow so that you can then search for "similar" ones?
On the example of news aggregators, where messages are combined into stories.
Preferably in Python.
Now I use the spaCy and Textacy libraries to highlight keywords. Then I look for intersections and sort in ascending order of the total weight of matching keywords.

key_terms = textacy.keyterms.key_terms_from_semantic_network(doc,
                                                                             normalize=normalize,
                                                                             window_width=15,
                                                                             n_keyterms=30)

But the accuracy is very lame