Teams that monitor ESG controversies usually have the opposite of an information shortage. A single incident can generate dozens of articles within a few days, each covering the same underlying event, often repeating the same facts with a few new details.  At a certain point, the sheer number of articles makes it hard to tell which developments are material and which are just the same story told again.

The volume is the part that breaks traditional approaches. Millions of articles are written every day across hundreds of languages, more than any team of analysts could read, let alone reconcile into a clear timeline of an evolving controversy. This is not a problem you solve by adding more people; the scale is on a different order of magnitude from human reading speed.

What changed is that language models can now read, categorize, and evaluate. They cover that volume in every language, judging whether two articles describe the same incident, whether one marks a new development, and how incidents link into a single controversy over time.  Leveraging the latest AI models is the only way to structure this much material and generate daily updates. 

SESAMm runs this across the ten million documents it ingests each day, from more than four million sources in over 100 languages, including premium news wires, NGO bulletins, company communications, and discussion forums. The result is ESG controversies organized into three layers: articles, events, and cases.

From Articles to Events to Cases

Each layer builds on the one below it, and each answers a different question an analyst needs answered.

Articles are individual news articles or documents: the raw material.

Events group the articles that describe the same specific incident or development. When forty outlets cover the same supplier labor issue, those forty articles become a single event, with the underlying coverage attached. Articles published close together in time and describing the same development are grouped; an article describing a genuinely new development, even on the same broader topic, forms a separate event. A strike in 2022 and a similar strike in 2024 at the same supplier are recorded as two events, because they are distinct incidents rather than a continuation of one.

Cases sit above events. A case ties together the events that belong to the same underlying controversy as it unfolds, with no fixed time limit. An oil spill, the regulatory investigation that follows it, and the settlement that closes it months or years later are three separate events but one case.

Articles tell you what was written, events tell you what happened, and cases tell you how a controversy is developing. All three sit in the same view: one entry per controversy, with the chronology of events nested inside it and the source articles a click below that.

Why the Underlying Data Matters

A three-layer structure is only as good as the data underneath it. To capture a controversy from start to finish, that data has to include the early signals that appear in regional press, NGO bulletins, or non-English sources before larger outlets report them, sometimes days later.

SESAMm's coverage spans more than 100 languages and extends well beyond mainstream news wires, so its cases are built on a wider base than most monitoring platforms screen. A controversy that starts in a local-language outlet, moves through regional media, and reaches the international press is captured as a single continuous case, rather than surfacing as disconnected alerts or being missed altogether in its early stages.

What Does This Change in Practice?

Three things change in day-to-day work.

The count starts to mean something. A rise in the number of cases reflects new controversies emerging, not an old one being picked up by more outlets.

Trajectories become visible. As a case accumulates new events over the months, the progression from complaint to investigation to hearing to settlement is easy to follow, rather than being buried in hundreds or even thousands of articles. 

Analysts spend their time differently. Less of it goes to clearing duplicate headlines, and more to the important judgment calls. 

What This Looks Like in the SESAMm Dashboard

In the dashboard, a company appears as a single entity with its related cases listed beneath it. Each case includes a controversy summary, an ESG risk classification, and an intensity score, with related events nested underneath and the original source articles just a click away. A case that draws on hundreds of articles becomes a short, readable list instead of hundreds of separate incidents.

Every case is fully traceable. Analysts can drill from a case down to its events, and from any event to the articles that produced it. The time period is set from the top of the dashboard, so older incidents do not crowd the view when the focus is on recent activity.

Reducing Noise in Adverse Media Monitoring

In practice, those forty articles collapse into one event, and that event sits inside a single case that is still developing, caught early and drawn from sources most platforms never see.

Grouping articles into events removes duplication caused when many outlets cover the same incident. Grouping events into cases keeps a controversy intact as it develops, rather than scattering it across months of separate alerts. Because this runs across ten million documents a day in more than a hundred languages, it holds up even for controversies that start far from the mainstream press.

The result is a view where the numbers carry meaning, the direction of an issue is clear, and the underlying articles stay one click away for full validation. 

‍

Researching and analyzing investment opportunities can be challenging for asset management—private equity and hedge fund portfolio managers, researchers, and analysts—because, of course, you want to make sure that you're a good steward of your client's investments.

And when you find and source data, such as traditional or alternative data, you also want to make sure it's reliable and that the methods used to gather it are tried and true.

This article aims to give you an inside look into SESAMm's knowledge graph—one of the key reasons SESAMm's NLP-derived alternative data is reliable and trusted. We'll explain what a knowledge graph is, why it's important, how it works, and what makes SESAMm's knowledge graph unique.

What is a knowledge graph?

A knowledge graph is a digital representation of a network of real-world entities, the foundation of a search engine or question-answering service. This structured data model puts the schema in context through linking and semantic metadata, providing a framework for data integration, analytics, unification, and sharing. In other words, it's like a map and legend, with the legend labeling the concepts, entities, and events and the map connecting and identifying their relationships. These details are stored in a graph database and visualized as a graph representation, hence the term knowledge graph.

Fun fact: The expression, knowledge graph, gained popularity after Google used it in 2012 to name their semantic network.

Two types of knowledge graphs

There are two general types of knowledge graphs: open and private. Open knowledge graphs are open to the public. They're created and made available by organizations such as Wikidata, DBpedia, and Yago. Private knowledge graphs are often only used by organizations that create them, like Google, WolframAlpha, Facebook, and SESAMm (of course). Some offer them up for a fee or subscription, such as Crunchbase and OpenCorporates.

Why a knowledge graph is important

Knowledge graphs are important because they equip us with a model to see how everything relates from a big-picture view, creating new knowledge. Its benefits include:

1. Incorporating disparate data sources, avoiding data silos
2. Integrating structured and unstructured data
3. Revealing insights from hierarchical data
4. Outlining relationships
5. Defining communities

Knowledge graphs inform machine learning algorithms

From a data science and artificial intelligence (AI) perspective, knowledge graphs provide machine-readable details, adding context and depth to data-driven AI techniques such as machine learning. Using knowledge graphs and machine learning models together improves system accuracy and extends the range of machine learning capabilities for better explainability and trustworthiness.

How a knowledge graph works

The core of a knowledge graph is its knowledge model, a collection of interconnected descriptions of concepts, entities, events, and relationships known as an ontology. This model provides a framework for statements or taxonomy. Each statement consists of a subject, predicate, and object (Figure 1)—known as a triple model—and each subject or object is represented only once in the context of the other subjects and their relationships. For example, in this simple sentence, "The boy kicks the ball," The boy is the subject, and kicker is the predicate because he kicks the ball, the object.

Likewise, each statement consists of three components: nodes, edges, and labels. A node, or vertice, represents an entity, which can be anything existing in the real world, such as a person, company, or object. For instance, in this example (Figure 2), Barack Obama is the subject node, Malia and Sasha are object nodes, and the edges, or relationships, are labeled as father or sibling, respectively.

What makes SESAMm's knowledge graph unique?

SESAMm uses open and private datasets with custom, curated information to create our proprietary knowledge graph. As a result, the knowledge graph is a vast map connecting and integrating over 70 million related entities and their keywords, relating each organization to its brands, products, associated executives, names, nicknames, and exchange identifiers in the case of public companies from a data repository made up of more than 18 billion articles and messages and growing.

The knowledge graph is updated regularly

Entities within the knowledge graph are updated weekly and tagged to ensure we correctly track their changes. For instance, the CEO of a company today might not be its CEO tomorrow. And brands might be bought and sold, changing the parent company with each sale. So, weekly updates within the knowledge graph ensure the system is aware of these changes.

NLP-driven accuracy

At SESAMm, named entity disambiguation (NED), a natural language processing (NLP) technique, identifies named entities based on their context and usage. Text referencing "Elon," for example, could refer indirectly to Tesla through its CEO or to a university in North Carolina. Only the context allows us to differentiate, and NED considers that context when classifying entities. This method is superior to simple pattern matching, which limits the number of possible matches, requires frequent manual adjustments, and can't distinguish homophones.

SESAMm uses three other NLP tools to identify entities and create actionable insights: lemmatization, embeddings, and similarity. The lemmatization process normalizes a word into its base form (morphology) to help identify and aggregate entities. Embedding assigns the entity a numerical value to help analyze how words change meaning depending on context and understand the subtle differences between words that refer to the same concept. Similarity measures whether two words, sentences, or objects are close to one another in meaning.

Learn more: “Gain Insights Fom Financial and ESG Data Using AI: A Comprehensive Guide.”

How SESAMm's knowledge graph benefits you

SESAMm tailored its knowledge graph to find, extract, and analyze data about public or private entities, which isn't readily available from the web or standard rating firms. This unique implementation of a knowledge graph provides insights to give you an edge when researching, analyzing, and submitting recommendations to the portfolio manager or clients.

SESAMm's premiere platform, TextReveal®, allows you to leverage NLP-driven insights fully and receive high-quality results through data streams, modular API and dashboard visualization, and signals and alerts. It's perfect for many quantitative, quantamental, and ESG investment use cases.

Learn how SESAMm can support you in your investment decision-making and request a demo today.

In recent years, the field of natural language processing (NLP) has seen significant advancements that have enabled more effective processing and analysis of large volumes of textual data. This has had major and disrupting implications for many industries, including finance and investment.

One area where NLP has been particularly useful is in the design of baskets and indices on all asset classes. One of them, where we can definitely observe key value-added, is related to digital assets and cryptocurrencies. A crypto basket is a group of cryptocurrencies that are bundled together and traded as a single unit, while a crypto index is a measurement of the overall performance of a group of cryptocurrencies.

Traditionally, the process of designing a crypto basket or index involved manually selecting a group of cryptocurrencies based on various criteria such as market capitalization, trading volume, and price history. However, this process is time-consuming and can be subject to bias. Moreover, it can be difficult for investors to invest in the crypto market as a whole. NLP technology has allowed it to analyze vast amounts of data from various sources such as news, articles, social media posts, and blogs to identify trends and sentiment around specific cryptocurrencies. This information can then be used to inform the design of crypto baskets and indices, making them more accurate and reflective of market sentiment.

By leveraging sentiment analysis through NLP-based indicators, robust indices can be created to serve as market benchmarks and investment vehicles. These indices can provide relevant performance measurement tools, allowing investors to understand the performance of their investments better and make more informed decisions. Furthermore, using NLP-based indicators to design crypto baskets and indices can also help generate alpha compared to a single basket of tokens. By tracking sentiment and emerging trends over time, investment professionals can gain valuable insights into which cryptocurrencies will likely perform well in the future and which may be less favorable.

Overall, using NLP in the design of crypto baskets and indices has significant potential to improve the accuracy and reliability of these investment products. By leveraging the power of NLP to analyze large volumes of text data, investment professionals can gain valuable insights into market sentiment and emerging trends, allowing them to make more informed investment decisions and potentially generate alpha. This is why we have entered a collaboration with Compass FT to design the first AI & NLP crypto sentiment index.

Index objective

The Compass SESAMm Crypto Sentiment Index aims to give investors exposure to the crypto market with a sentiment tilt to determine the selection and weights of underlying tokens. The index selects tokens based on financial filters such as average trading volume and market capitalization. Using NLP-based sentiment scores in the weighting mechanism allows the index to rebalance towards the coins with the best sentiment scores efficiently and, therefore, those with the highest expected relative returns.

Key features

• Provides smart and dynamic exposure to the cryptocurrency market
  
• Monthly review to adapt to the fast-moving crypto ecosystem and capture up-to-date/representative sentiment for each coin
• Unique quantitative weightings mechanism based on liquidity filters and sentiment scores
• Invests in a basket composed of the 20 main crypto coins
• Constituents are selected based on rigorous criteria considering liquidity, tokenomics, sentiment, custody, and security
• Methodology and governance in line with the most constraining financial indices regulation, the European Benchmark Regulation (EU BMR)

Index mechanisms

The Compass SESAMm Crypto-Sentiment Index is a diversified digital asset index designed to offer broad exposure to the market’s top crypto assets (all sectors included) while capping each component exposure at 30%. Weightings are based on sentiment scores, liquidity, and market capitalization constraints.

SESAMm’s NLP technology carries out a granular and transparent analysis of publicly available articles. More than 20 billion articles from over 4 million international and local sources are analyzed to identify each coin’s associated mentions. For each source, indicators of sentiment and volume of mentions are determined. These indicators are then aggregated daily to create a historical time series per cryptocurrency, which acts as the basis for the overall score used by Compass Financial Technologies. For each day and each coin, SESAMm calculates crypto sentiment scores based on several indicators, such as polarity, volume, and memory functions, to provide up-to-date and representative scores. SESAMm’s Crypto sentiment scores are based on the sentiment scores (negative, positive, and neutral) computed on articles related to the 50 digital assets universe.

Analytics

Reach out to SESAMm

TextReveal's web data analysis of over five million public and private companies is essential for keeping tabs on ESG investment risks. To learn more about how you can analyze web data or request a demo, contact one of our representatives.