Introducing the Corporate Sentiment Tracker

| Andrew Duchon


The Corporate Sentiment Tracker™ is based on data derived from Manzama’s Insights™ platform.    Insights is a patent-pending system that uses machine learning to classify corporate news into 6 factors and 25 subfactors (e.g., news related to a company’s products, their labor force or deals with other companies).   Within each subfactor it will also classify whether the news is positive or negative for the company, from which we can derive a Corporate Health Score which ranges from -10 to +10 with most companies having a score close to zero, that is, the news is fairly balanced.   Insights also classifies news as to whether it is expressing an opinion or not.  

An additional functionality of Insights is to group articles together into stories and storylines.  A story is composed of all the articles from different sources published that day that concern the same event.   Storylines group stories across days, so one can see how an event evolves, or very similar types of events over time are recurring.  Based on the article count, or number of different sources all talking about the same thing, we can describe some stories as being “bigger” than others. 

The Corporate Sentiment Tracker looks at a subset of the outputs of this system, namely articles in the Management subfactor which express an opinion.   In general, these are articles where the CEO, or other named person from the company, is expressing an opinion about their company, industry, or economy in general.  Articles where the CEO is talking about a product, or other subfactor, will generally be classified into that respective subfactor.   The goal is to capture management’s opinions about the state of the world in general, now and into the future.  Occasionally, you may see articles that are opinions about the CEO, not by the CEO.  These are rare and should not have a major impact on the values seen on the website. 

Insights ingests news data from over 56,000 English-language sources.  The Tracker is using news concerning 7,500 public companies around the world, but due to the language restriction, Management opinions are more likely to show up from US, Canadian, UK, and Australian companies.   About 125,000 articles per day come through the system, but just a fraction of these will be part of the tracker.   

We cannot guarantee that every management opinion article is captured, or that it is classified “correctly” (and humans will also have different thoughts on what is correct).   Thus, it is best to think of this as a sampling method, with a confidence interval that we estimate at +/- 5%. 

In general, we think of this “good” and “bad” news as a valence: positive, negative, or neutral and we try to be objective about the event from the company’s point of view.   Opening new stores is clearly positive, closing stores is negative.   In the case of Management Opinion however, we can think of these designations as an actual sentiment, a feeling, expressed by a person.    

Starting from the top of the Sentiment Tracker page, we will describe each component: 

  1. Corporate Sentiment over Time
  2. Current Sentiment
  3. Recent Stories
  4. ESG Trending Topics
  5. Top Story for Each Topic
  6. Trending Terms
  7. Top Story for Each Term

Corporate Sentiment over Time 

At the top is the Corporate Sentiment over Time.   The percentages shown are the percent of companies, of those companies that week that had a management opinion article, where that opinion was positive.   The equation is simply:  

Most weeks there are about 100 companies that have this type of news.  The older the data, the fewer companies there are likely to be in the score, as we are still processing and reviewing the historical data. 

Current Sentiment 

The Current Sentiment is this week’s current Percent Positive.  The week begins on Monday (UTC) so as the weekdays unfold, the score may change, more drastically from Monday to Tuesday.   The score is updated a few times throughout the day with new stories and the last update time is shown. 

Recent Stories 

The Recent Stories section shows Management Opinion stories sorted by the most recent date and then by article count. 

Trending Terms 

The Insights system automatically extracts terms (up to 5 words long) from the headlines.   The top terms will mostly be single words however.   These are ranked by how many different companies use that term in the Management Opinion stories.    

The health scores shown in the timeline are based on the article count that day (week or month) of stories using that term.  The scores can range from -10 to +10.  It will be more positive if there are many more articles using that term in a positive story than articles in negative stories.   Even a term like “pandemic” can show up in positive stories, e.g., the CEO thinks they are recovering well from the pandemic. 

Top Story for Each Term 

For each of the most frequently used terms, we show the top story, which is the largest by article count using that term in the given time range.   This is present merely to show an example of how that term has been used. 

ESG Trending Topics 

For an Environment, Social and Governance (ESG) framework, we use the World Economic Forum’s International Business Council (IBC) ESG metrics outlined in their recent whitepaper: “Measuring Stakeholder Capitalism Towards Common Metrics and Consistent Reporting of Sustainable Value Creation”.   The topics have been expanded slightly for better differentiation, e.g., Biorisk has been added to capture news and risks related to the Coronavirus (and other biological outbreaks affecting corporations, e.g., listeria).   News is classified into a topic if it contains at least one term related to that topic, e.g., Biorisk is a top topic recently since it captures all the different ways it might be discussed (“Covid-19”, “Covid”, “Coronavirus”, “SARS-CoV-2”, etc.). 

Top Story for Each Topic 

For each of the most frequently discussed topics, we show the top story, which is the largest by article count and topic relevance in the given time range.   This is present merely to show an example of how that topic has been discussed. 


The score was 80% yesterday, but is 70% today…what happened? 

Each week, starting Monday, the companies with Management Opinion stories are accumulated.  By chance, on Monday, most of the stories were positive.  On Tuesday, more stories came out and more of these happened to be negative.  So, throughout the week, the percent positive could change radically.   The timeline shows the final score that week (Monday through Sunday). 

You said that once the week is over, the score is set, but I saw a historical week’s score actually change from what it was last week. 

We are constantly reviewing the outputs of the system and revising which articles are considered management, what their valence is, and which are expressing opinions.    

What’s up with Elon Musk? 

Celebrity CEOs like Elon Musk could make the news for saying anything about anything.  We try to put these headlines into the neutral valence, since they often have little to do with the company or the economy.  However, since almost anything can happen, they may still show up, especially as opinions about the CEO.  However, since the Sentiment Tracker is based on the proportion of companies, all these stories will just count as one data point, or even wash out with both positive and negative stories. 

That person is not in management, shouldn’t you exclude them? 

You will occasionally see statements by economists from a large bank, or others from a company.   These tend to be useful for assessing the economy.  We do keep track of who CEOs are at different companies, but other personnel maybe sometimes slip in.  Again, this is part of the sampling error. 

I know the CEO of company X said this thing about shareholder activism, but it’s not showing up in the tracker.  What’s the deal?  Or, CEO’s are not talking about this term in the tracker, so they must not care about it? 

The tracker is part of a larger system which is classifying news into many categories, only one of which shows up here.   So, that CEO’s statement might have been classified into some other category, e.g., Shareholders.  When there are statements about something more specific about their company, e.g., a CEO talking about the release of a new product, or how they are expanding into a new territory, how they are helping their employees, then those articles will likely be put into those categories.   The “Executive” news the tracker is using is when they are making more general statements that can’t be put anywhere else; they must also be classified as an opinion and not a statement of fact.  So, just because it does not show up in the tracker does not mean it wasn’t captured somewhere else in the larger system. 

About the Author

Andrew Duchon

Director of Data Science at Manzama, a Diligent Brand

Dr. Andrew Duchon is the Director of Data Science at Manzama, Inc. There, he has developed and continues to improve the machine learning system behind Manzama’s patent-pending Insights™, which automatically assesses company news for indicators of corporate health. Prior to Manzama, Dr. Duchon was a data scientist at Jama Software building models to predict customer churn. For eight years he worked at Aptima, a defense industry consultancy, as a project manager, researcher, and software engineer applying natural language processing to understand team behavior in a variety of domains and has four patents related to this work. Prior to Aptima, Dr. Duchon was a co-founder and Director of Technology at, a semantic search engine. He has also worked as a private consultant developing statistical models for the finance industry; an NSF postdoctoral fellow at Brown University studying cognitive maps in virtual reality; and a researcher at the Advanced Telecommunications Research Institute in Japan developing neural networks for speech recognition. Dr. Duchon holds a Ph.D. in Cognitive Science from Brown University, and a B.A. in Psychology and the Integrated Science Program from Northwestern University.