I have lots of short text messages and sentences, most of them don't have a common context (and sense). And they are bi-lingual (in one phrase can be words types in English and other (Cyrillic) language). And I need to understand somehow which of the sentences is 'having sense' and not a random combination of words.
And I have a huge list of words, like a white-list (let's count it has N length) by which I can determine that phrase or text is useful and so on.
If the phrase is having two or three+ words from the white list it is determined as sensetive. (Actually there isn't just one list, but three of them, with it's own priority and so on.
Of course, there are various ways to find these words in JS/TS text phrases one-by-one.
For example, forming sentences as an array of words and checking them via .include or check string via Regexp. But I guess NER is a perfect choice for this problem.
I tried to use DeepPavolov for this, but as far as I understand it has pre-build words in its vocabulary. But in this case, I should train my model and I am badly familiar with python.
On the other hand, I have libraries like NLP.js (badly typed) and compromise but I never use them and don't know how resourceful they could be.
And I guess there should be some kind of online NER solution, in which I could build my vocabulary online via UI.
So my question is: should I use any kind of NER solution at all? And if you have any advice, I'll be glad to receive it. Or there is any other option available, which could solve my problem? I am dealing with typescript/javascript and node.js server-side. So I have enough time and resources for analyzing, but I can't deal with python and Jupiter Notebook and other ML things, since I have no huge experience with them.