Customer Profiling / Pattern recognition

[anon33357744]

I have a question about profiling.
I need to analyze a dataset that involves the following case study:
In the Trade Finance area of a bank we have transaction data in Excel. The Excel file contains information such as transaction ID, booking country, type of trade (Export, Guarantee or Import), Beneficiary, applicant, Goods, Amount, Currency, Beneficiary country, applicant country, port of origin, port of destination, origin of goods, client risk and country risk.

The question here is whether you can have a kind of 360 degree customer view with the help of this data or what you can do with this data. I thought of pattern recognition in which I visualize the relationship between Beneficiary and applicant. If our customer has always sent his contractual partner the same type of products on average or the price has moved within a certain level, a kind of detection should take place, which can then tell me whether something is wrong with the transaction. The same applies if you have sent the goods to the contract partner for years via the same route and then suddenly use a different shipping route. Is this use case possible in KNIME?

I already started to prepare the data, but have a question to “Strings to Document”. The column “Beneficiary” is filled with a company name as Siemens AG and only the column “goods” is filled with different ways of description like “boys pullover”, “chemicals Pharm” or “baby goods” and also longer descriptions. Should I apply “Strings to document” just for the column goods?
If not, how can I use a Strings to document node for more than one column?

Thank you for your ideas and help

cheers,
Canan

[Kathrin]

Hi Canan,

That sounds like in interesting project!

One idea is to consider your task as an anomaly detection project. You could try out two things:

1. Analyse the data and extract rules. Maybe also somehow automated using a series of ETL node (GroupBy, Pivoting, Rule Engine…).
2. Train a predictive model (e.g. Random Forest) for each feature based on the other features in your dataset. Then apply those models to new transactions. If the values differ from your prediction you could classify it as a strange transaction.

In this white paper we describe something similar for sensor data. It is not the same as in your case, but maybe it helps to get an idea about anomaly detection.

181212_Whitepaper_Anomaly_Detection_Predictive_Maintenance_KNIME.pdf

About your question regarding the Strings to Document node. This node is a node from our Text Mining extension. It converts different string columns (e.g. Title, Autor, Full Text) into a “document”, which is a special data type in KNIME for text mining. If you want to know more about text mining I can recommend the book “from words to wisdom”, but as far as I understand your problem you don’t need to transform your strings into documents.

Please let me know in case you have any questions.

Cheers,
Kathrin