Learn more about our AI and Machine Learning technologies and how we use them to bring you our 20+ influencer profile metrics.
How Does Influencity Gather Data?
Influencity harnesses the power of Artificial Intelligence (AI) to understand the content of an influencer’s posts and interactions. We don’t rely on just one data point, such as facial or image recognition, we take into account a wide range of public data such as bios, captions, comments, location tags and hashtags and train the machine learning that’s behind the Influencity platform to detect demographics, language, and even fraudulent accounts.
Influencity data comes from external databases. We use both public sources, such as the Spanish Statistical Office (INE), the United States Census Bureau, and the Mexican National Institute of Statistics, Geography and Informatics (INEGI), as well as private ones. By bringing all of this information together and cross-checking all available data points, Influencity provides clients with a full picture of:
- The location of an influencer and their audience.
- Influencer and audience demographics, such as age range and gender.
- The quality of an influencer’s followers.
- Audience ethnicity.
- Other influencer and audience information such as language spoken, interests and brand affinity.
Note: Private information can only be read with an influencer’s consent. For more information on this, please see How Does Influencity Respect Privacy Laws?
How Do We Detect an Influencer’s Location?
To identify an influencer and their audience's location, we use the following:
Location Tags in Instagram Posts
This applies not just to cities, but also to restaurants, museums and important landmarks in each city. For example, an image of a meal tagged at a trendy restaurant in Paris will be associated with this city. For TikTok and YouTube, we used location tags to pinpoint the specific country.
Natural Language Processing (NLP)
This cutting-edge technology helps computers scan and interpret human language. Using this allows Influencity to “read” all public data from an influencer’s profile bio, YouTube About section, captions, comments and hashtags. Here are some examples of what NLP can do:
- Read the text in Instagram and TikTok profile bios and YouTube About sections - This is the free text space in a profile where influencers can describe themselves. Often, they will include their location in this section and Influencity detects this using NLP.
- Hashtags - We can read and sort hashtags related to countries, and, in the case of Instagram, cities. This includes hashtags from restaurants, landmarks, points of interest, etc.
This branch of AI allows computers to “see” images the way humans would and then classify them according to their content. For example, if an influencer uploads a photo with the Eiffel tower in the background, this photo will be associated with Paris.
Weighted Data Points
Influencity gathers all of these data points above and triangulates them to identify an influencer’s location with the highest precision possible. In addition, some data points are weighted more heavily, according to their relevance.
To give you an example, an influencer may have ‘Paris’ in their Instagram bio or YouTube About section. This data point would be weighted more heavily than just one picture with the location tagged as Paris. Similarly, six months of posts with location tags in various restaurants and museums in Paris would have more weight than just one post with a Paris #hashtag. This is how we differentiate between an influencer who actually lives in a city and one who is just visiting.
How Do We Detect Influencer and Audience Demographics?
With the help of facial and image recognition, Influencity can “view” the content of images to detect influencer and audience gender and age range. Just as humans scan profiles and recognize an influencer’s gender and approximate age, our technology can understand the content of an image and sort it accordingly.
In addition, Influencity uses NLP to interpret the text in all public bios, posts and comments. We take into account data points such as an influencer’s @handle or YouTube channel name (whether it’s a male or female name) and the pronouns (his, her, they) associated with a particular influencer profile.
The same goes for age. For example, if an influencer uploads a photo with a caption such as “celebrating my 25th birthday in Miami,” NLP would read this just as a human would and associate this influencer with the 25-34 age range.
As with the location metric, Influencity carefully triangulates data to reduce the margin of error. For instance, facial recognition may have one margin of error – like if a profile doesn’t have many pictures of the user’s face. However, if this influencer has the name Amy in their bio and handle, then this margin of error diminishes. The more photos and data points we include, the more this margin of error is reduced.
How Do We Detect Fake Profiles or Bots?
To understand how we detect fake profiles and bots, it’s first important to understand how they work. Bots are created to mimic human activity online and are usually produced on a large scale. Since bots are created in large batches, they tend to have the same programming, and thus, the same online behavior.
Just like a human might see a fake account and find it suspicious, Influencity AI can detect these patterns through reverse engineering and use this information to identify fake accounts.
Here are some of the things our AI will look for:
- Profile information - A fake account might not have a profile picture or could have a strange name, such as a series of numbers and letters.
- Followers to following ratio - Bots favor mass actions and will follow tons of profiles. They may all follow each other and have few authentic followers of their own.
- The number of posts - Similar to the point above, a group of bots may all have the same 3 photos on their profile and will like each other’s posts. While they may be “active” online, they’re not regularly uploading authentic content as a real user would.
- Comment authenticity - We use NLP to run semantic and syntax analyses of a post’s comments. Suspicious and repetitive comment patterns are associated with a low follower quality.
- Spam links - A fake profile may have links to spam websites in its bio, captions and comments.
Want to know more? Read all about our Follower Quality Metric to see how it can help you identify influencers with the most authentic audience.
How Does Influencity Detect Inactive Accounts?
Unlike bots or fake profiles, inactive followers were created by real people, they may have just lost access to that account or have decided to stop using it. We can detect these accounts by analyzing their interactions with other users. If there has been no activity such as likes, comments or new posts in the past 6 months, this profile will be considered inactive and will not be useful to your influencer marketing strategy.
How Does Influencity Detect Audience Ethnicity?
We use face and image recognition to provide our users with aggregate data on audience ethnicity.
How Does Influencity Recognize Language?
We rely on a Natural Language Processing algorithm to analyze bio, post and comment data and identify the language used. The NLP algorithm is 100% accurate, making it easy to precisely filter your influencers and audience by language.
How Do We Identify Brand Affinity, Interests and Content Topics?
Influencity uses AI and NLP to interpret data from pictures, captions and hashtags and then classify content according to this information. If an influencer mentions a brand in a post or interacts with their page, this would count as brand affinity. Another example of this is our ability to identify logos; if an influencer uploads a photo holding a Starbucks mug with the logo clearly visible, we can identify this and this brand will be associated with this influencer on our platform.
The same goes for interests and content topics. If an influencer uploads a photo doing yoga, image recognition will understand this and the term ‘yoga’ will then be added to the interests and topics related to that profile. If an influencer is posting plates of food and tagging restaurants, they would be categorized as a foodie. Learn more about our Interest filter and how to use it here.
How Accurate is Influencity Data?
Our data is based on the triangulation of around 100 different data points. By constantly cross-checking our information, we’ve guaranteed a minimum 95% accuracy rate on each metric, achieving 99% accuracy on many of them and 100% on a few, such as gender.
How Does Influencity Respect Privacy Laws?
Our platform gathers aggregate data and doesn’t associate specific identifiable information with an influencer. That is to say, while we might know the age and interests of an influencer based in London, we don’t associate the name of this influencer with this information.
We only use identifiable data in these three cases:
- When there is a legitimate interest.
- When Influencity acts as a data sub-processor on behalf of our customers.
- When we have an influencer’s expressed consent.