Page 17 - Contributed Paper Session (CPS) - Volume 6

CPS1465 Claude Macchi et al.
concepts and associate companies with an economic activity code. This allows
a reduction of the interpretation of texts describing the activities of businesses
and facilitate the attribution of codes. The system is built in stages, like an
onion, layer after layer. In the first step, companies are coded only at the
aggregate level of 2-digits of the NOGA classification. A codification at the
most detailed level of 6-digit will be undertaken at a later stage, once the
system has made its first experiments and has reached a sufficient level of
stability and quality.
NOGAuto is based on the principle of feedback, correction and continuous
improvement. Each action can be challenged and thus trigger the relaunch of
one or more steps in the process. This approach is fundamental not only for
the construction phase, but must also be continued once the system will be in
production.
In machine learning projects, communication is fundamental from the
beginning. Any change often causes resistance, and especially the word
"automation" is easily linked by the staff to "loss of jobs". It is therefore
essential to integrate future users of the system into the project from the
beginning and involve them in the development. "Automation" should in no
way be seen as a reduction of tasks, but as a chance to be able to use the time
saved for new tasks. The NOGAuto system is not limited to the codification of
economic activities of enterprises, but can be adapted and used in the context
of codification based on classifications in any other field of statistics.

6 | I S I W S C 2 0 1 9