Should I use NLP to detect entities in metadata? How?

Question

I have some meta-data about a building, here is one of the example:
AHU-S-6F-01.RA.CO2.1
I am finding a way to make the computer able to recognize entities in metadata like:
[Location].[Sensor-Type].[Sensor-Element].[Sensor-ID]
There have other data formats in the dataset, so I am thinking I can use Named Entity Recognition (NER) and implement it by Tensorflow.js to make the computer learn to recognize entities in metadata and give a tag to them. I found an example below:

https://monkeylearn.com/text-analysis/

Is Named Entity Recognition(NER) suitable to use to detect entities in metadata? How can I implement it by Tensorflow.js?
If not, what should I use to solve this problem?
Thanks!!!

Answer 1

If your info is quite predictable (deterministic), where each field is always expected and each value is predefined, I would go for regexp.

NLP/NER technology requires some training (hundreds to thousands of training examples) and should the values change over time, you have to retrain with the new examples. It can handle situations where ambiguities arise (maybe not your situation).

Answer 2

you need to detect which words are product features so you should use BERT with this archirecture to match complicated patterns ( but if your problem is simple,it's better to use regex)

this architecture implemented in similar tasks like these:

https://gab41.lab41.org/how-to-fine-tune-bert-for-named-entity-recognition-2257b5e5ce7e

https://towardsdatascience.com/named-entity-recognition-ner-with-bert-in-spark-nlp-874df20d1d77