AI in Instagram
Published by Sid Chadha on September 2nd, 2020
Instagram has used Artificial Intelligence to become the top of the social media game. As we all use Instagram, its increasingly important to understand the algorithm behind these. Below are statistics to provide context on how popular and widespread Instagram has become.
Display of instagram's statistics, Hotsuite.com
Artificial Intelligence is used in multiple ways in Instagram:
​
Explore Page and Search Function
The explore page is the cornerstone of Instagram and thus makes it critical to describe this process. AI in the explore page works through a recommendation system. Through using a process called word embedding, Instagram's AI algorithm studies the order of words with a caption or message. When a certain word in the caption is ranking more popular, Instagram will pair that with another account that has been liking accounts with captions/messages with similar words. Specifically, a similar post is sorted into a group of 500 for that user and the AI filters this group on spam, misinformation, and policy violation content. In a blog post by Instagram, they described this process through having three passes:
​
​
​
​
​
​
​
​
​
​
​
​
​
​
The 25 leftover posts will be posted on your explore page as a result.
Instagram's AI learns more about what to recommend as you interact with the explore section of instagram further. For example, if you saw a post about science and liked it and another post about sports and ignored it, science will be given a higher "weight" in the algorithm and sports will be given a significantly lower "weight". People can make their message and caption broader to reach a broader audience.
​
They additionally give a lower weight to a certain author or seed account after already seeing a post from that same account so you don't keep seeing the same post or same posts from a certain user. The weight keeps getting lower as you see more posts. Instagram additionally uses Bayesian optimization tool to tune the value model efficiently and often as users continue to user Instagram and Instagram continue to update their platform. Below is a visual of how the explore page works:
How the explore page could work; Facebook
How the explore page works; Facebook
Target Advertising
​
Instagram extracts insights from the customer, such as search preferences, engagement insights, and more, to sell data to companies who can then target their advertising to certain profiles. They use Artificial Intelligence in order to help companies advertise to certain profiles. This could be built through a deep forward neural network with the data process being extremely challenging with over 1 billion users and thus an immense amount of data to analyze.
​
Target Advertising illustration; later.com
Enhance the User Experience
​
Instagram has to ensure that users find value in their platform and to do that, they must find content that the user finds relevant. Artificial Intelligence is the most efficient way to do exactly that. Instagram has changed their feed structure from reverse chronological order to displaying posts that the user would like and share with an algorithm similar to the explore page. The algorithm learns over time what the user likes most through its interactions and then creates a personalized feed through this data (deep forward neural network)
​
Filter Spam
​
There are often fake messages from accounts and spam bots. Instagram uses past data from what it has detected as spam and uses artificial intelligence to automatically remove "spam" posts and comments. Through using Facebook's AI algorithm DeepText, Instagram is able to easily understand each message and its context accurately.
​
Fight Cyberbullying and Delete Offensive Comments
​
With Instagram being one of the largest used platforms to report bullying (where in a survey 42% of more than 10,000 UK youth between ages 12 and 25 reported Instagram was the platform where they were most bullied), Instagram has become the first company to use Machine Learning to remove offensive comments and posts. In contrast, most other social media platforms, like Twitter or Facebook, used to have humans look at the actual text and delete it. Instagram uses a neural network called DeepText and by having humans create a dataset with what offensive comments look like, the algorithm can then predict what is offensive and remove it.
​
Study the Human Condition
​
In a select study, 100 million Instagram photos were used to learn global clothing patterns. This immense amount of data is extracted and analyzed with AI. Through extracting insights about clothing, Instagram is able to study, social, cultural, and economic factors. Through using Convolutional Neural Networks to analyze the images and deep forward neural networks or Recursive Neural Networks to predict socioeconomic background, humungous datasets are able to be formed and analyzed.
From filtering spam to creating an optimal explore page, AI has helped instagram gain its enormous following and become one of the top social media platforms to date.
​
​
Acknowledgements​
https://blog.hootsuite.com/instagram-statistics/
https://bernardmarr.com/default.asp?contentID=1396
https://ai.facebook.com/blog/powered-by-ai-instagrams-explore-recommender-system/