RepRisk is uniquely positioned to take advantage of the latest advancements in natural language processing, a branch of artificial intelligence focused on giving computers the ability to understand text.
The latest and most important developments in natural language processing have been made with pre-trained transformer-based language and generative text models.
Bidirectional Encoder Representations from Transformers (BERT) models process natural language and identify context and meaning through an attention mechanism that mimics human cognitive abilities by highlighting relevant parts of the input data in a forward and backward direction simultaneously. RepRisk uses 17+ years of human-labeled data – a training set of 1 million documents – to refine the latest BERT models to specialize in environmental, social, and governance (ESG) and business conduct risk text analyses in 23 languages. The models are finetuned with input from our multilingual human analysts, ensuring that as the landscape and language of ESG evolves, our models continue to capture the most relevant risk incidents.
Text generation models learn from existing text data, creating new content by rearranging and combining established patterns. RepRisk employs the latest models to support human analysts in writing risk summaries for captured risk incidents with increased speed and efficiency.
RepRisk has the right data and the right use case because of our unique combination of artificial + human intelligence:
◾ The right data: highly accurate human-labeled data used for training and refining models in a feedback loop.
◾ The right use case: on a daily basis, scraping huge amounts of unstructured data from public sources and stakeholders.
Source: RepRisk ESG data science and quantitative solutions, www.reprisk.com
# The RepRisk ESG risk incident journey from source to database
Source: RepRisk ESG data science and quantitative solutions, www.reprisk.com
From vast amounts of data – 2,500,000+ documents screened daily from 150,000 sources in 23 languages – RepRisk uses machine learning models to identify and classify ESG-related risk incidents consistent with definitions from key international standards and norms. Documents that contain relevant ESG and business conduct risk information are appended with machine learning predictions and submitted for human analysis according to a proprietary logic that ensures the most relevant information is reviewed first.
Our team of 150+ human analysts supervises the artificial intelligence process and provides a deeper analysis of risk incidents, including assessing severity, reach, and novelty of each incident and providing further review in instances where, for example:
- the ESG criticism is weak;
- there is no direct link to a particular company; or
- the source is making a political statement
Our team of analysts then integrates generative text models into the process of writing risk incident briefs, utilizing the models for speed and efficiency. Importantly, every risk incident is reviewed by a human analyst before being published to our client-facing solutions.
Our artificial + human intelligence approach results in the world’s largest and most comprehensive database on ESG and business conduct risks.
Copyright 2022 RepRisk AG. All rights reserved. RepRisk AG owns all intellectual property rights to this methodology and to the content of this website. This information herein is given in summary form and does not purport to be complete. Any reference to or distribution of this methodology must include the entire methodology to provide sufficient context. The information provided on this website does not constitute an offer or quote for our services or a recommendation regarding any investment or other business decision. Should you wish to obtain a quote for our services, please contact us.