Question: usecase „order confirmation“

Hi guys,

I’ve recently saw some videos about deep-learning-neuronal-networks-things and then I’ve found your node-based-software and now I wonder, if we can use all this “crazy stuff” to do something nifty.

I think, a “usecase” that is very common, easy to explain, easy to understand, but still impressive would be an interesting start.

So my suggestion is an automated “order confirmation” workflow.

If one company wants something from on other company, there will be a quotation (Angebot), an order (Bestellung) and if the order corresponds with the quotation there will be an order confirmation (Auftragsbestätigung).

So we have two inputs (quotation and order) and one output (order confirmation).

The inputs will be PDF-Files, sometimes text-based (e.g. printed from word document) sometimes picture-based (e.g. a scanned document).

The “body” of the quotation will be:
1. The address of the customer
2. The address of the quoting company
3. The quotation number
4. A table with the columns (e.g. Position number, Amount, Product/Service, Price) and rows with numbers and text.

The “body” of the order will be:
1. The address of the quoting company
2. The address of the customer
3. The quotation number
4. The order number
5. A table with the columns (e.g. Position number, Amount, Product/Service, Price) and rows with numbers and text.

Now comes the (for humans) tedious part. You have to check if the addresses from the quotation corresponds with the addresses from the order, then you have to figure out what amount of Produts/Services they have been ordered and if the price was correctly calculated.

If everything is OK, there will be an order confirmation.

The “body” of the order confirmation will be:
1. The address of the customer
2. The address of the quoting company
3. The quotation number
4. The order number
5. A table with the columns (e.g. Position number, Amount, Product/Service, Price) and rows with numbers and text.

The output should be a text file with all the information from the order confirmation body.

At this point, I’ve no idea if this is a possible workflow for your program.

If it is possible, I would like to know how I can train the neuronal network?

Lets say, I have 500 corresponding quotation->order->confirmation example sets, can I “feed” them to the neuronal-network (nn) and the nn will figure out by themselfe what I want from him or do I still have to do all the “enrichment” on the input data?

It would be great to hear what you think of this usecase!

Kind regards

Matthias

Hi Matthias,

You should be able to parse the pdf files with the Apache Tika integration available with KNIME Analytics Platform. Have you already tried it out? Later, some of the tasks that you want to implement might be computed using the Rule Engine node, by comparing two or more string columns. Finally, you can write back a text file, by using the CSV Writer node for example.

I am sorry but I did not get what you want the NN to do. What would be the prediction that you have in mind? If you want to train an algorithm using texts, you should first convert them into documents, do all the pre-processing steps, transform the documents into a vector representation that can be interpreted by an algorithm and then you can train the algorithm.

Hope that helps,

Best,

Vincenzo