tobias-kirschstein / greenspectors Goto Github PK

Extract the information from the ESG reports and structure them in a way s.t. we can use them for ML models/ranking

• Parse word or extract otherwise
• Find categories or other ways to structure the underlying data
• Rank companies based on ESG fact sheets

• Find suitable pretrained model for sentiment analysis
• Evaluate how well model works on our dataset
• Rank companies based on twitter tweets

Pretrained Sentiment Analysis Models:

• Flair: https://medium.com/@b.terryjack/nlp-pre-trained-sentiment-analysis-1eb52a9d742c
• BERT fine-tuned on Twitter Sentiment Analysis: https://towardsdatascience.com/sentiment-analysis-in-10-minutes-with-bert-and-hugging-face-294e8a04b671
• Transformers Library: https://towardsdatascience.com/sentiment-analysis-with-pretrained-transformers-using-pytorch-420bbc1a48cd
• TweetBERT (not released?): https://arxiv.org/pdf/2010.11091.pdf
• VADER Sentiment Lexicon (proabably quite simple): https://pub.towardsai.net/ai-sustainability-tweets-sentiment-analysis-using-pre-trained-models-8669f8ba70ed

Benchmarks/Datasets:

• Sentiment140: http://help.sentiment140.com/home, https://www.kaggle.com/kazanova/sentiment140
• ACL 14 Twitter Dataset: https://www.researchgate.net/figure/Experimental-datasets-based-on-SemEval-2014-and-ACL-2014-Twitter_tbl1_339964194
• SemEval-2017 Task 4: Sentiment Analysis in Twitter: https://aclanthology.org/S17-2088.pdf

• Find out which actions get a lot of attention on Twitter
• Analyze if some actions always/never/sometimes lead to positive perception

• market growth of greenliness consulting sector
• Investor interest in green companies (ESG funds etc.)
• Catchy name
• Business model canvas
• Persona

• If there are discrepancies find out why
• Kind of like PCA for the measures to see which ones have more positive impact

• Find suitable Twitter datasets
• Think of how to filter tweets (Company + Sustainability)
• Filter out crazy tweets
• Weight tweets by number of likes/retweets/followers ...

• Repository structure
• Create example notebook for letting stuff run on Azure

How should the final product look like? How will it be used?
A tool to help the CEOs & company management identify the most impactful green measures
Just a search field for measures?
Rating of companies?
Something else?

• Redeem Azure Voucher
• Make GPUs available