You may be wondering about the relevance of a machine learning system integrated into your AP automation solution. Then, this article is for you.
The two possible approaches
There are 2 approaches to locating information on a document:
Geographic analysis based on the position of the data on the invoice.
Semantic analysis: data detection by text analysis and detection of keywords, data formats, etc.
Self-learning involves memorizing a position when the user clicks on an area, thus allowing an existing model to be corrected. However, how do you know if this new position is final or temporary?
This self-learning technology by geographic area is obviously part of the panel of in-house technologies available to us.
However, we have chosen to completely deactivate these old self-learning functions in Yooz because we want to eradicate any notion of the geographical position of a data (not sustainable, specific to a supplier) to favor semantic analysis (perennial and independent of the supplier).
In addition, it is important to keep in mind that solutions that offer self-learning are necessarily in one of the following scenarios:
either they do not have a shared knowledge base, and in this case the user has very tedious manual learning work to do upstream for the recognition to work,
or they have a shared database, and in this case, if a user makes an error during the zoning, self-learning is wrong and at the same time penalizes all the other customers.
Neither of these two cases seems convincing to us.
The Yooz approach
Yooz's technology is called "full-text", which means that it involves absolutely no prior configuration (no zoning, no invoice templates, no self-learning). The entire invoice is read and recognized the first time without the user having to set anything.
Our technical guideline is very clear on this point: we do not want to define a field by its geographical position, in particular because this position could change if the layout of the invoice were to change.
However, we are, of course, in a process of continuous improvement. As soon as a user detects an invoice that might require recognition optimization, they can report it to us with a single click from the Yooz platform. This invoice will then be the subject of optimization work by our specialists. This optimization, always generic, is based on full-text semantic analysis algorithms.
Then, this development will benefit all Yooz customers.