Demonstrate ESG scoring

[MichaelTiemannOSC]

Databricks provides the following benchmark for developing ESG scores using NLP:

https://databricks.com/blog/2020/07/10/a-data-driven-approach-to-environmental-social-and-governance.html

We should chart how this can be accomplished with Jupyter Notebooks using resources assembled in the Data Commons platform. If we decide to do this, let's build a proper set of milestones and subtasks to track progress and completion.

@toki8 @HeatherAck

[ChristianMeyndt]

@MichaelTiemannOSC This classification of ESG topics reminds me of what Michael Platt pesented to us some time ago. Is he still in the development team of OS-C?

[ChristianMeyndt]

@HeatherAck @JeremyGohBNP @LeaADeleris @andraNew @OferHarari @idemir-ids @DaBeIDS @mriefer Here are some additional ideas for NLP solutions. We should discuss if we also add this to our planning for next year.

[OferHarari]

Thanks, @MichaelTiemannOSC, and @ChristianMeyndt. The approach makes sense in high level. However, it has many issues.

• this is all topical and not quantitative metrics that can be used to have a real apple to apple comparison, or to be viewed as viable by a regulator.
• topics + sentiment (as described) is a good approach. However, it is challenging to reach high precision rates.
• Correlation between entities is also prone to many errors. Even in the example above - they did not address the fact that nasdaq appears highest , likely because of bias in news where every news mentioned nasdaq as an exchange and not as nasdaq the entity. Facebook and Reuters are the easy example. Amazon and Apple will appear in many stories but with low relevance.

Factset Truvaluelabs is attempting to provide "objective ESG scores mined from news". Worth looking if they are willing to contribute some data.

If we go this direction - we should:

• focus on very few topics
• proper annotation (negative and positive examples)
• metrics that might resonate with the regulator (to become mandatory further down the road)

and present something of high quality that users will view as valuable.