Twitter has an entity resolution problem

Elon Musk has recently been asking his Twitter followers what he should do with Twitter if he is successful in acquiring the company. I think Twitter has a major entity resolution problem, and sorting it should be one of Elon’s priorities.

Have you ever blocked someone annoying on Twitter or Instagram, only to see them pop up again a few days later with a fresh account, following you and pestering you again? The same happens when trolls get banned from an entire platform. They often reappear with a fresh account.

That is an entity resolution problem. It’s the same person — it’s just that these social media platforms do not realise, and therefore treat them like a fresh, new, innocent account.

It’s a potentially dangerous problem. We block people because they are annoying, we block them because they might be scammers, but we also might block them because they are abusive or potentially even our stalkers.

Blocking someone on a social media platform should mean that they are blocked even when they set up a new account. It’s the person (or bot!) you really want to block, not the account.

Currently, Twitter and Instagram try to handle new account creation in different ways.

To sign-up to Twitter you need either an email address, or a mobile phone number. Each email address, or phone number, can only be used for one Twitter account. Try to sign-up again with that email address or phone number, and you will be told it is already in use and that you must use an alternative.

So if you are a social media pest, it is pretty simple to just create a completely new Twitter or Instagram account under a new email address. So what could these social media giants do to prevent these bad actors operating with freedom with new accounts on their platforms?

Well, users potentially share more information about themselves than just their email address or phone number. Using this data (which is often messy and inconsistent) to associate two accounts together, is essentially the science of entity resolution.

In fact, it’s no different to fraud detection entity resolution that payment and eCommerce giants must employ to avoid losing money to criminal fraudsters. If social media platforms treated new account creation as seriously as Buy Now Pay Later (BNPL) companies do, then they might be able to keep more of the bad actors out.

Interestingly, Twitter’s latest Terms of Service update, which will go into effect on June 10, tells us the data they collect, which could be used for entity resolution to connect together user accounts. This includes:

Inferred Identity. We may collect or receive information that we use to infer your identity as detailed below:
When you sign in to Twitter on a browser or device, we will associate that browser or device with your account. Subject to your settings, we may also associate your account with browsers or devices other than those you use to sign in to Twitter (or associate your signed-out device or browser with other browsers or devices or Twitter-generated identifiers).
When you provide other information to Twitter, including an email address or phone number, we associate that information with your Twitter account. Subject to your settings, we may also use this information in order to infer other information about your identity, for example by associating your account with hashes of email addresses that share common components with the email address you have provided to Twitter.
When you access Twitter and are not signed in, we may infer your identity based on the information we collect.

So maybe Twitter is planning to do more to keep track of the bad guys, and stop them signing up multiple times to their platform. However, we know that the biggest technical challenge in doing this successfully is ingesting entity data in real-time, whilst simultaneously searching that data and also being able to search complex entities quickly. At Twitter’s scale this is all the more challenging, but it certainly is possible with the right entity resolution solution.

At the time of writing, it is still unclear whether Elon Musk’s bid for Twitter will be successful. Nevertheless, one of his stated priorities has been to sort out Twitter’s bot problem, where entity resolution based account matching, combined with AI-based detection of non-human behaviour, might be key. Regardless of who is holding the majority of Twitter’s shares in the next few weeks, this should be a priority for the company to re-establish trust.