We Ran the Mueller Report Through Our AI and Here is What We Found…
Background
- On April 18, 2019 the Department of Justice (DoJ) released the Special Counsel Robert Mueller’s report on Russian Collusion in the 2016 US Elections (Volume I) and Obstruction of Justice (Volume II) by the president of the United States, Donald J. Trump.
- According to the PDF association the document is an image only PDF that was scanned and compressed at least once and more probably twice, with redactions being added to the first scanned document using software.
- Because of the format, the document is unsearchable and difficult to consume, with significant implications for journalists, attorneys, and others who wish to quickly draw insights from the text.
- To remove this unnecessary burden, we performed an NLP sentiment analysis of the 448 page report using our AI platform.
What did we want to find out?
When we started thinking about this problem, one of the questions we had was: What is the general sentiment found in the language of the Mueller Report with respect to the most relevant players that were talked about in the report? To that goal, we processed the entirety of the Mueller Report through our NLPbased AI platform to derive important insights present in the report that may otherwise be too time consuming or difficult to infer. In this article, we shall focus on the sentiment analysis of specific people mentioned in the report.
Results
Number of Mentions and Sentiment Variability
Shown below are the number of mentions for specific keywords (entities) throughout the report. Out of interest we selected the following key individuals as keywords: “Trump”, “Putin” and “Clinton”.
Not surprisingly, Trump has the most mentions out of the three individuals and, along with Clinton, is mentioned a significant number of times in both volumes of the report.
On the other hand, Putin has very few mentions in Volume II, which makes sense given that Volume I deals with potential collusion with Russia and Volume II deals with obstruction of justice charges.
Turning to sentiment analysis of the same keywords, we find that the “Trump” keyword is generally mentioned in a positive light. Conversely, we find that the general sentiment around the keyword “Clinton” is negative for most of the report, especially during Volume I.
Focusing on Volume I, the keyword “Putin” is mentioned in a positive light during the initial and final sections of Volume I, but in a negative light in the middle sections of Volume I of the Mueller Report.
Next, we show the distribution of overall sentiment for several key individuals discussed in Volume I of the Mueller Report. The size of the bubble indicates the number of mentions for that particular keyword and the color indicates the sentiment value.
We find that Stone, Papadopoulos, Clinton, Gates, Podesta and Muellerare mentioned with an overall negative sentiment. Whereas, Trump, Cohen,Manafort, Comey and Sessions are mentioned with an overall positivesentiment. Flynn is the only individual who is mentioned neither in a positive or negative way overall.
In Volume II the overall sentiment was negative for Sessions, Clinton, Mueller, Cohen and Rosenstein and positive for McGahn, Gates, Trump, Priebus and Manafort. Interestingly, in contrast with the results from Volume I, in Volume II Flynn is not talked about in a negative light. Overall, discussions around Comey had a neutral sentiment.
Paragraph Sentiment
Below we show paragraphs extracted from the report that were associated with negative sentiment in the case of Clinton and positive sentiment in the case of Trump.
Section Sentiment
We can also compute the average sentiment of particular sections of the report where certain keywords are mentioned. For this analysis, we considered both Volume I and Volume II with the keywords Trump and Comey, shown in the two plots below, respectively.
Each bubble is a section of the report and the size of the bubble is a proxy for the number of mentions of the keyword in that section. The farther the bubble is to the right, the longer the section is. Average sentiment is shown on the y-axis with negative sentiment towards the bottom and positive sentiment towards the top.
In the plot below we highlighted one of the bubbles corresponding to section “2. The President Seeks to Have McGahn Dispute the Press Reports” which shows that in that section there were six mentions of Trump and the average sentiment associated with that keyword was negative.
In the following plot, we show that in section “4. The President Sends Messages of Support to Cohen” of the report, the keyword Comey had an average positive sentiment with 37 mentions.
It is also interesting to note the distribution of the sentiment for different sections of the report in the context of specific keywords. In the cases shown above, it is clear that Comey is for the most part discussed in a positive light. For Trump the results are more mixed but as a whole still seem to be on the positive side.
This is just a small display of the power that the AI wields and does not do justice to the full capabilities of our platform. If you are interested please contact us at human@commerce.ai and we would be happy to demo our platform or the analysis of the Mueller Report itself.
Disclaimer:
The Commerce AI platform is typically used for analyzing large unstructured data about consumer products. We ran this report through our AI engine as an experiment to find out the kind of insights it surfaces. This is purely a data science experiment and does not signify any political affiliation for Commerce AI or its team members. All images, except for the word cloud, are property of Commerce AI and are direct screenshots from our dashboard.
There are limitations with our approach, particularly because our platform is specifically designed for the retail and commerce ecosystem and because the nature of the report is to be objective and therefore devoid or lacking emotion. Hence, one must not conclude this is solely how Mueller and his legal team felt personally about the individuals mentioned in the report. It is also a measure of the overall sentiment about these individuals mentioned in the numerous quotes found in the report.
