Advice on NLP


It’s my first post here, and I’ve just finished Part 1 in a rush, so I might say something dumb :).

I’m struggling with the following problem and I’m trying to see which is the best approach:
I work in a construction company and it takes a lot of time “translating” the customer needs into the our internal classification of products so we can prepare a cost estimate.

We usually get the request for quotation in an Excel file, and we’ve seen just about every possible variation for 2 by 4 wood beam we can think of. So now we go trough 2-300 lines of excel, and try to understand what the customer needs, and then go and find it in our database so we can add the costs for it.

Should I look into the NLP course, or should I use other techniques?
I’ve looked around, and it seems that the problem is a variation of product matching used in e-commerce sites.

And to make things worse, I’m from Romania, and the requests are obviously in Romanian :slight_smile:


Can you post a representative example of one (preferably translated to english) to give me a better understanding of inputs and outputs? No need for real customer data if you would prefer to make a dummy example, but just a simple example so we can get an idea of what the data going in looks like, and what the desired prediction would be would be helpful.


It seems the English phrase for what I’m looking for is Bill of Quantity.
Here is a small youtube video explaining what it is, and how it looks like:


Hmm. I may not be the best person to answer this as I don’t have a ton of experience with this, but I will share my thoughts in case it’s helpful. It doesn’t look like the actual materials are always specified explicitly in the work, otherwise you could look for having it pull the materials out of the text and label them as “materials” (kind of like amazon comprehend if you want to look up an example). Since it doesn’t look like they are all specified in the work explicitly, i would think it would be a multi-label NLP classifier to take the text and label it with all the materials that are needed for that work.

My thought for getting quantities would be to then have an additional model that would take materials, text and quantities to do a regression to predict the qty of each material. I would treat the model to get quantity as a second model that takes the data and the output classes from the first model as inputs. There may be a way to do this all in one model, but that’d be what I would shoot for as I am a relative newbie.