Sber has created an open-source NLP model recognizing and classifying texts in 47
environmental, social, and governance topics.
The Artificial Intelligence Laboratory and the ESG Risk Office at Sberbank designed
ESGify, a user-friendly NLP model to evaluate corporate ESG risks, First Deputy
Chairman of the Sberbank Executive Board Alexander Vedyakhin said at AI Journey
2023.
The NLP (Natural Language Processing) model processes and structurizes texts in
Russian and English. A dataset of 2,500 unique texts about ESG risks was used to train
the model, followed by a two-stage verification of the dataset markup results to improve
the quality.
Drawing on a risk-oriented approach, the classifier can correlate any text from public
sources with a specific type of realized risk or indicate its non-present. This is how it helps
analyze ESG risks based on public information when there is no standardized data about
companies.
The service is embeddable in any ESG risk assessment product or mechanism. ESG risk
classification helps to drive ESG practices in data science and business communities.
Sber’s AI model can help assess the investment appeal of a business taking ESG risks
into account, and help in screening suppliers, counterparties and customers for
compliance with ESG principles. The service will also be useful to consulting firms and
analytical agencies.
Alexander Vedyakhin, first deputy chairman of the Executive Board, Sberbank:
“Sber was one of the first in Russia to introduce ESG principles into all aspects of its
businesses and has been sharing its expertise, innovative tools and services with
businesses and regions for several years to facilitate their ESG transformation.
Our AI model, which automatically identifies risk types, helps optimize ESG risk work. ESGify draws on an advanced hierarchy: the model is able to categorize texts into 47
environmental, social and governance topics. It is continuously being trained and
developed. Moving forward, the service is expected to be able to identify risk significance
in addition to their classification.”