In this section, we present a case study about the company ExxonMobil as a representative company of the energy industry by performing an end-to-end process for the ESG news related to ExxonMobil.

Firstly, we gathered the data needed for the analysis by developing crawlers that accumulate data into our datasets from alternative sources and company filing reports. As sources for the alternative data, we use Google RSS news feed, GDELT, and Yahoo Finance API. We collect specific news that is related to targeted ESG topics. We extract the news based on co-occurrence between the company we analyze and the topics we predefine (ex. Exxon climate change). We store the news title, content, date, and domain site with search query metadata. The company documents are gathered through the portal of the U.S. Securities and Exchange Commission (the “SEC”). For the filings, we store the whole document.

We continue filtering and preprocessing the news when we have all the data. In our batch streaming, we often encounter news that is irrelevant to our topics or company. Therefore, we filter the news by the frequency of appearances of our customized taxonomy words. Based on a predefined minimum word occurrence threshold, we classify the company news as ESG related or not. After filtering the news, we calculate the sentiment of the news by using a predefined Hugging Face pipeline, “sentiment-analysis.” The pipeline creates its embeddings, and afterward, leveraging the transformer architecture, it calculates a number between -1 and 1, indicating the sentiment(-1 is negative, 1 is positive). For our purposes, we define a threshold for the certainty of the sentiment, hence, for every value above 0.7, we classify the news as positive, and for every sentiment value below -0.7, we classify it as negative.

For a better insight into the results, we created several charts for the data collected between 05/01/2022 and 06/15/2022. Using the frequency of appearances of our taxonomy words, we mapped the values on the chart and colored them with their sentiment. The line indicates the number of times the term was found in the news content. The line’s color indicates the sentiment where red is negative and green is positive. We repeated this for the three ESG categories: environmental, social, and corporate governance. 

In Figure 1, we observe that ExxonMobil has more news about negative community behavior than positive, which indicates that their social index will fall. On the other hand, they provide health care for their workers and are involved in positive human rights movements.

 In the following bar plot, Figure 6, we observe the number of appearances of the term ESG in negative and positive news. The red bar indicates the number of occurrences of the word “ESG” in the week’s negative news related to ExxonMobil, starting with the date on the x-axis. The same applies to the green bar with the difference in sentiment.

Our following analysis involves Relation Extraction, which helps us extract meaningful connections between entities from unprocessed texts and use the relations to create a Knowledge Base. Using this method, we generate a Knowledge Graph from the text in Def 14A, a proxy document reported by ExxonMobil for 2022. The model has detected that, in their proxy statement (Def 14A), ExxonMobil specified that they are founders of an organization called Alliance to End Plastic Waste, the graph below. 

To compare this graph with the news, we create a knowledge graph after crawling the web for news with the query “Exxon Alliance to end Plastic Waste”, the graph below. As we can see in the Figure, we detect a potential greenwashing. We find a connection between a parent organization American Chemistry Council connected with relation’ instance of’ with Lobbying.

To ensure our results, we continued our research by creating one more Knowledge Graph from the news results of the search query “American Chemistry Council Lobbying,” graph below. We find that the news that resulted in the relation between American Chemistry Council and Lobbying has a headline: “Oil-backed trade group is lobbying the Trump administration to push plastics across Africa.” With this, we can conclude that they are doing some kind of lobbying so they can proceed with using plastic products which is both violation of environmental issues and a corporate governance problem.

When considering ESG investments, portfolio managers look into companies or funds with high ESG ratings. Several companies, including MSCI, S&P, Sustainalytics, CDP, ISS, Bloomberg, and others, provide ESG ratings. However, the consensus level regarding the ratings is relatively low, which makes ESG-focused investments challenging. For example, let’s look at the correlations of ESG Ratings across different providers. We observe a correlation of 7% between ISS and CDP ratings, 16% between MSCI and CDP, 58% between Sustainalytics and Bloomberg, and the highest correlation of approximately 74% between Bloomberg and S&P. This dispersion among ESG ratings from different data providers is confusing for investors. The discrepancies observed in the Figure below occur because many of the ESG measures are subjective, disclosure can be inconsistent, there may be biases in the data, and there could be changes in corporate behavior.

Existing solutions from technology-based providers offer insights into how companies can improve their ESG index or score and get a better overall view of their index of sustainability. The insights are generated by building a specific model for the company based on accessing, organizing, and analyzing relevant company files. Another interdisciplinary, analytical solution to increasing the ESG transparency is to provide datasets of raw ESG indicators. This solution involves collecting company data from various sources, such as company documents, news, social media, and alternative data sources. The data can then be used either in a row format or in a more insightful way based on collaborative work between data scientists and high-level investor knowledge of market insights and company ESG involvement. This model offers excellent synergetic and profitable cooperation for influencing corporate behavior and contributing to enhanced sustainability and financial stability.

Machine learning and AI-based solutions can contribute to the meaningful representation of each company’s most critical ESG factors, emphasizing corporate dedication to ESG issues by creating dashboards showing what the company reports and what the company does about ESG. The dashboard would extract and visualize the overall picture of a company’s ESG involvement. The insight is generated by collecting every data point for the company from internal and external sources and giving more profound insight into the company sustainability index. The AI-based solution cold model dynamic changes and go beyond ESG reporting by providing a future estimated outlook into corporate ESG risks. On the other hand, the AI-based solution could offer positive impacts indices and explainable results from machine learning analysis and extract insights about how social and environmental trends will impact the company. A similar study could offer insights into how negative scenarios (e.g., climate or social disasters) will affect the company’s future and determine its resilience to adverse events.