How to Interpret Hamilton 2.0
Hamilton 2.0 displays outputs from sources that we can directly attribute to the Russian, Chinese, or Iranian governments or their various news and information channels. These channels and accounts often engage with topics, hashtags, URLs, and people that are in no way affiliated with the Russian, Chinese, or Iranian government. It would therefore be INCORRECT to, without further analysis, label anyone or anything that appears on the dashboard as being connected to state-backed propaganda.
Note – The dashboard uses natural language processing and other machine learning and auto-translation tools to extract key information from the data. Although highly accurate, automated systems are imperfect. We urge all those reporting on the dashboard to confirm findings with ASD.
How often is the dashboard updated?
The Twitter section of the dashboard updates several times per day, meaning that recent Tweets may or may not be included in the dashboard’s data. Website, YouTube, and MFA data is updated on a daily basis. The default setting displays data from the previous seven days, but users can filter data by any date range.
Twitter Data
What accounts are included?
The dashboard displays aggregate results from accounts connected to the Russian, Chinese, or Iranian government, diplomatic corps, and state-funded media. Accounts are filtered by country and category (diplomatic/government accounts and media accounts). Click here for the full account list.
What accounts are included in each category?
The Hamilton 2.0 dashboard monitors roughly 1000 accounts representing individuals or entities connected to the Russian, Chinese, or Iranian government or Russian, Chinese, or Iranian state-funded media. The dashboard collects data on accounts that primarily target foreign audiences, namely embassies, consulates, foreign ministries, ambassadors, key government figures, and international media outlets and their affiliated channels, programs, and key personnel. We identified these accounts using basic open source research techniques, including but not limited to Twitter’s state-affiliated media and government official labels. Most of them openly note their affiliation with the Russian, Chinese, or Iranian government or state-funded media (i.e., these accounts represent overt rather than covert influencers). We audit each list on a monthly basis to add relevant accounts and to remove those that are defunct or no longer relevant.
We divided Russian, Chinese, and Iranian accounts into two broad categories: diplomatic/government accounts and state-funded media accounts. A detailed description of each category is below:
Russian, Chinese, and Iranian diplomatic/government accounts
This category contains accounts of key individuals and institutions that represent the government of the Russian Federation, the People’s Republic of China, and the Islamic Republic of Iran, including:
- All known embassy, consular, ambassadorial, and consul general accounts as well as influential diplomats (those with more than 5,000 followers)
- Official accounts representing the Ministry of Foreign Affairs, as well as agencies and organizations that have an international or foreign policy focus (e.g., Russian Mission to NATO)
- Key individuals within the Russian, Chinese, or Iranian government
Russian, Chinese, and Iranian state-funded media accounts
This category contains accounts representing Russia, China, and Iran’s various state-funded media outlets, including:
- Accounts connected to Russia, China, or Iran’s state-funded international broadcasters and online news and information portals
- Accounts connected to television programs on state-backed media outlets
- Accounts of key management personnel
- Accounts of influential reporters, journalists, and show hosts affiliated with state media outlets that Twitter has labelled as state-affiliated media, that explicitly reference employment with a state-affiliated outlet/program, or that have more than 5,000 followers.
Note – The media account list does NOT include accounts connected to individual journalists, reporters, or television personalities who contribute to or are associated with Russian, Chinese, or Iranian state-funded media outlets but are not notable accounts (defined by less than 5,000 followers) or are not regularly featured in state media.
Top 10 accounts
These charts display the most active, retweeted, and liked accounts during a specified time period. This information is useful in interpreting results on the rest of the dashboard. For example, the most active accounts will affect the top 10 results more than accounts that Tweet infrequently, regardless of the popularity of those accounts. Engagement metrics are determined using Twitter’s public API and are updated during each refresh for tweets posted within the last seven days — historical tweets (those older than seven days) are updated on a rolling basis. Because the dashboard does not update in real-time, we urge users to check up-to-date engagement metrics on Twitter.com.
Top 10 tweets
This chart displays the Tweets made by monitored accounts that have received the most engagement over a specified time period. Non-English Tweets are displayed in English using Microsoft’s Translator API. Engagement metrics are determined using Twitter’s public API. As noted above, tweets within the last seven days are updated during each refresh — historical tweets (those older than seven days) are updated on a rolling basis.
Top 10 hashtags
This chart displays the hashtags (#) most frequently used by monitored accounts during a specified time period. Hashtags are useful indicators of topics of interest, but hashtags can be used either to support or undermine the specified cause or person. They therefore should be interpreted in the full context of the Tweet(s).
Top 10 mentions
This chart displays Twitter handles most frequently included in Tweets made by monitored accounts. These results include all mentions of (also known as tags), replies to, and retweets of a specific account. Accounts that are monitored on the dashboard (e.g., @RT_com) are filtered out, and therefore are NOT included.
WARNING: This data does not necessarily imply that the Russian, Chinese, or Iranian government is supporting or targeting a mentioned user.
Top 10 key phrases
This chart displays the most-used phrases (including proper names, terms, etc.) that appear in Tweets made by monitored accounts during a specified time period. See below to learn how we extract key phrases from Tweets.
How are key phrases extracted from Tweets?
We use Microsoft’s Cognitive Services to translate non-English Tweets and to extract key phrases. Non-relevant or non-specific key phrases (e.g., verbs, common nouns, etc.) are filtered out to remove noise from the results. We extract key phrases instead of key words to link together related words (e.g., “INF Treaty” or “Barack Obama”). It is possible, however, for two or more related nouns to appear as multiple entries in search results. For example, “Trump,” “Donald Trump,” and “President Trump” may appear as three unique phrases. Searching for a term in the “key phrases” section of the FILTERS sidebar will reveal similar key phrases.
Top 10 links
This chart displays the hyperlinks most shared by monitored accounts in a specified time period. Shortened links have been expanded to display the full URL.
WARNING: The dashboard does not screen links for malware, phishing content, spam, or any other undesired content. Click on the links at your own risk.
Top 10 mentioned countries
Using key phrase extraction, this chart displays the countries most mentioned in Tweets during a specified time period. In addition, special territories and regions (e.g., Hong Kong and the EU) are included as individual entities in the country data. The chart lists all mentions of a country, including alternative country names, under that country’s official name (e.g., mentions of US, USA, U.S., and America are all included in tabulations for the United States of America).
Note: Mentions of the State of Georgia will likely be mistaken for the country of Georgia. As with all key phrase extraction, other false positives are possible.
Top 10 Tweets retweeted by monitored accounts
This chart displays the most popular Tweets (as determined by retweet metrics) that have been retweeted during a specified time period by at least one monitored account. The default setting displays retweets of accounts NOT monitored on the dashboard. The account that posted the original tweet is displayed in the chart; the monitored account(s) that retweeted that tweet is/are displayed in the Top 10 Accounts chart. The purpose of this chart is to display the messages and, occasionally, messengers being amplified by official Russian, Chinese, or Iranian accounts.
WARNING: A retweet does not necessarily imply that the Russian, Chinese, or Iranian government endorses the author or the content of a Tweet.
How are key phrases extracted from Tweets?
We use Microsoft’s Cognitive Services to translate non-English Tweets and to extract key phrases. Non-relevant or non-specific key phrases (e.g., verbs, common nouns, etc.) are filtered out to remove noise from the results. We extract key phrases instead of key words to link together related words (e.g., “INF Treaty” or “Barack Obama”). It is possible, however, for two or more related nouns to appear as multiple entries in search results. For example, “Trump,” “Donald Trump,” and “President Trump” may appear as three unique phrases. Searching for a term in the “key phrases” section of the FILTERS sidebar will reveal similar key phrases.
Broadcast Data
Summaries of RT America/RT UK and CCTV+/CGTN America’s YouTube content
The dashboard summarizes the outputs of video content uploaded to RT America and RT UK’s YouTube channels (on the Russia dashboard) and CCTV+ and CGTN America’s channels (on the China dashboard). NOTE: YouTube videos over 20 minutes in length are NOT included. Each video segment represents a single YouTube clip. Data is provided by our partners at VidRovr. To learn more about VidRovr click here.
The titles and descriptions associated with YouTube clips are those provided by the channels. The tags are derived from information provided in the chyrons or through VidRovr’s computer vision algorithms, including key contextual information and the names and titles of interviewees and those appearing on-screen. Human coders determine the countries and categories associated with clips.
How does ASD collect this data?
Data is provided by Vidrovr, an enterprise video understanding platform. Vidrovr leverages proprietary machine learning and computer vision algorithms to determine who is speaking and appearing on screen, the key topics that are associated with a specific section, and what is appearing visually, among other things. For any inquiries concerning Vidrovr, please send an email to contact@vidrovr.com
Hamilton 2.0 uses VidRovr technology to extract the entities and text that appear in each segment and to automate coding suggestions. Human reviewers manually review automated suggestions, but they do not view each video and therefore will not necessarily identify all key elements of each video.
How often is the dashboard updated?
The broadcast section of the dashboard updates daily, meaning that broadcasts from the current day will not be included on the dashboard until the following day. The default setting displays data from the previous seven days, but users can filter data by any date range.
Due to the fact that human coders must categorize content, there occasionally will be one- or two-day lapses in manual coding, particularly during weekends and holiday periods. In addition, disruptions in video feeds will occasionally result in days when video data is not captured. YouTube data will be backfilled as soon as possible.
How are associated countries identified?
Country information is determined by scraping video descriptions and by human coders. Coders base their determination on the textual data in the chyrons and titles, and only label countries that are of primary or secondary importance. Coders can tag multiple countries associated with an individual clip. In addition, special territories and regions (e.g., Hong Kong and the EU) are included as individual entities in the country data.
How are key phrases extracted from videos?
We use Microsoft’s Cognitive Services to extract key phrases from titles and descriptions. Non-relevant or non-specific key phrases (e.g., verbs, common nouns, etc.) are filtered out to remove noise from the results. We extract key phrases instead of key words to link together related words (e.g., “INF Treaty” or “Barack Obama”). It is possible, however, for two or more related nouns to appear as multiple entries in search results. For example, “Trump,” “Donald Trump,” and “President Trump” may appear as three unique phrases. Searching for a term in the “key phrases” section of the FILTERS sidebar will reveal similar key phrases.
How are news categories coded?
For the first year of the dashboard, human coders categorized each segment based on the set of pre-determined categories and sub-categories listed below. After compiling a large data set of human coded samples, we used machine-learning and automation to determine the following categories:
- Business – Content related to specific business entities or a business sector not covered by one of the other categories
- Celebrity/Gossip – Content related to celebrity news and gossip
- Civil Disorder – Content related to protests, riots, strikes, etc.
- Climate Change/Environment – Content related to climate change, including global warming, sea-level rise, climate science, etc.
- Crime – Content related to violent crimes, missing persons, mass shootings, the criminal justice system, and other true crimes (not including police brutality – see Social Inequality)
- Culture – Content related to music, film, television, theater, pop-culture events, festivals, etc.
- Energy – Content related to the energy sector or energy politics/geopolitics
- Free Speech/Censorship – Content related to censorship, press freedoms, and alleged bias in the technology sector
- Geopolitics – Content related to foreign policy and international relations
- Health/Safety – Content related to public safety issues, health advances, health concerns, global pandemics, health care systems, etc.
- Human Interest – Content related to emotional coverage of a person, people, or pets
- Income Inequality/Poverty – Content related to the wealth distribution, poverty, homelessness, minimum wage issues, etc.
- International Economics/Trade – Content related to trade, the global economy, international sanctions, etc.
- Man-Made Disaster – Content related to plane/car crashes, explosions, fires, chemical spills, nuclear disasters and other human-induced calamitous events
- Media Criticism – Content related to claims of media bias and criticism of the “mainstream,” Western, or specific media outlets
- Migration/Immigration – Content related to migration or immigration issues
- Military/War: Content related to war and conflicts between and among states, and content related to military exercises, training, military-technological developments, etc.
- Natural Disaster – Content related to earthquakes, storms, floods, landslides, etc.
- National Economy – Content related to domestic economies and financial sectors
- Politics (U.S., U.K., Russia, China, Other) – Content related to domestic political affairs, elections, etc.
- Political Scandals – Corruption, sex scandals, political investigations, electoral fraud/ballot stuffing, abuses of power, etc.
- Racism/Discrimination/Inequality – Content related to issues pertaining to racism, xenophobia, LGBTQI rights, religious discrimination, social justice issues, etc.
- Religion – Content related to religion or religious studies, but excluding religious discrimination (see social issues/inequality)
- Science/Technology – Content related to STEM topics, emerging technologies, space exploration, and the tech sector (not including military technology – see Military/War)
- Sports – Content related to team or individual sports, excluding gossip or scandals related to individual athletes (see Celebrity-Gossip)
- Surveillance/Spying/Privacy – Content related to government or corporate abuses of privacy, including surveillance and spying
- Terrorism – Content related to domestic or global terrorism, including incidents, threats, groups, etc.
- Weather – Daily weather reports not related to natural disasters or climate change
The purpose of categorical coding is to assist users in finding relevant material. All coders are trained in order to ensure consistency; however, due to the volume of content and the real-time nature of the dashboard, inter-coder tests are not always possible. Thus, the potential for human error and variability is greater than would be acceptable in an academic study. We therefore urge caution when using or citing categorical data. If you would like to help enhance our ability to code and verify this data, please donate to ASD at this link.
Website Data
The dashboard displays aggregate results from English-language websites and news portals funded by the Russian, Chinese, or Iranian governments, and regional and different language versions of RT and Sputnik News. Currently, the Russia section of the dashboard collects data on English-language and regional versions of RT.com, Sputniknews.com, RBTH.com (Russian Beyond the Headlines), and TASS.com (Russia News Agency). Additionally, the dashboard collects data from websites that have been connected publicly to Russian intelligence, including Southfront, Newsfront, and Strategic Culture Foundation.The China section of the dashboard collects data on CGTN.com, Xinhuanet.com, Globaltimes.cn, en.people.cn (People’s Daily), and Chinadaily.com.cn. The Iranian section collects data on PressTV.com, enfarsnews.ir (Fars News Agency), and en.irna.ir (Islamic Republic News Agency). Data is provided by our partners at Debunk.EU, an initiative combating disinformation in Lithuania. Click here for more information about Debunk.EU.
Data sources
Data is extracted from article excerpts (first 200 words or 10 percent of the article) from more than 20 Russian state-funded websites, five Chinese state-funded websites, and three Iranian state-funded websites monitored on the dashboard. This means that the metrics provided on the dashboard SHOULD NOT be considered a full and complete summation. It is also important to stress that key phrases and countries not mentioned at the beginning of articles will not be captured by the dashboard. Thus, the results should be interpreted as an approximation of key messaging interests, rather than as a complete tally of all terms and phrases in collected articles.
How often is the dashboard updated?
Website information is updated once per day, meaning that articles from the current day will not be included on the dashboard. The default setting displays data from the previous seven days, but users can filter data by any date range.
Top 10 articles by social media engagement
This chart displays the articles that have received the most engagement on Facebook over a specified time period. Engagement numbers are from public pages only. Data is provided by Debunk.eu, and reflects accurate totals at the time the dashboard is updated each day, meaning that actual engagement numbers may vary depending on the time of day.
Top 10 key phrases
This chart displays the most-used phrases (including proper names, terms, etc.) that appear in article excerpts and titles during a specified time period. See below to learn how we extract key phrases from articles.
How are key phrases extracted from articles?
We use Microsoft’s Cognitive Services to extract key phrases from article excerpts and titles. Non-relevant or non-specific key phrases (e.g., verbs, common nouns, etc.) are filtered out to remove noise from the results. We extract key phrases instead of key words to link together related words (e.g., “INF Treaty” or “Barack Obama”). It is possible, however, for two or more related nouns to appear as multiple entries in search results. For example, “Trump,” “Donald Trump,” and “President Trump” may appear as three unique phrases. Searching for a term in the “key phrases” section of the FILTERS sidebar will reveal similar key phrases.
Top 10 countries
Using key phrase extraction (see above to learn how we extract key phrases), this chart displays the countries most mentioned in article excerpts during a specified time period. In addition, special territories and regions (e.g., Hong Kong and the EU) are included as individual entities in the country data.
NOTE – The chart lists all mentions of a country, including alternative country names, under that country’s official name (e.g., mentions of US, USA, U.S., and America are all included in tabulations for the United States of America).
MFA Statement Data
The dashboard displays aggregate results from official English-language press releases and statements published by the Russian, Chinese, and Iranian Ministries of Foreign Affairs and Permanent Missions to the United Nations.
Data sources
Data is extracted from statement excerpts (first 200 words) from the Russian, Chinese, and Iranian Permanent Missions to the United Nations. This means that the metrics provided on the dashboard SHOULD NOT be considered a full and complete summation. It is also important to stress that key phrases and countries not mentioned at the beginning of articles will not be captured by the dashboard. Thus, the results should be interpreted as an approximation of key messaging interests, rather than as a complete tally of all terms and phrases in collected statements.
How often is the dashboard updated?
UN and MFA statement information is updated once per day, meaning that statements from the current day will not be included on the dashboard. The default setting displays data from the previous seven days, but users can filter data by any date range.
Top 10 key phrases
This chart displays the most-used phrases (including proper names, terms, etc.) that appear in statement excerpts and titles during a specified time period. See below to learn how we extract key phrases from statements.
How are key phrases extracted from statements?
We use Microsoft’s Cognitive Services to extract key phrases from statement excerpts and titles. Non-relevant or non-specific key phrases (e.g., verbs, common nouns, etc.) are filtered out to remove noise from the results. We extract key phrases instead of key words to link together related words (e.g., “INF Treaty” or “Barack Obama”). It is possible, however, for two or more related nouns to appear as multiple entries in search results. For example, “Trump,” “Donald Trump,” and “President Trump” may appear as three unique phrases. Searching for a term in the “key phrases” section of the FILTERS sidebar will reveal similar key phrases.
Top 10 countries
Using key phrase extraction (see above to learn how we extract key phrases), this chart displays the countries most mentioned in statement excerpts and titles during a specified time period. In addition, special territories and regions (e.g., Hong Kong and the EU) are included as individual entities in the country data.
NOTE – The chart lists all mentions of a country, including alternative country names, under that country’s official name (e.g., mentions of US, USA, U.S., and America are all included in tabulations for the United States of America).