Twitter Has an Entity Resolution Problem

When Elon Musk raised the question of how many Twitter accounts are actually bots, he highlighted a problem that goes far beyond social media: how do you know which accounts belong to real, unique people?

This is an entity resolution problem — and it’s one that every platform with user-generated accounts faces.

The Bot Problem Is an Identity Problem

Bot networks don’t operate with completely random data. They reuse identity attributes — the same phone number across multiple accounts, similar email patterns, shared device fingerprints, IP addresses from the same subnets.

The challenge is detecting these connections at scale. Twitter has hundreds of millions of accounts. Checking each account individually against a ruleset misses the network effects. You need to look at the connections between accounts to find clusters of related identities.

This is exactly what entity resolution does.

How Entity Resolution Detects Bot Networks

An entity resolution system would approach the bot problem differently from traditional account-level fraud detection:

1. Ingest all account attributes — name, email, phone, device ID, IP, user-agent, creation timestamp, behavioral patterns
2. Resolve entities — find accounts that share attributes, even partially (fuzzy matching on names, pattern matching on emails like user1234@domain, user1235@domain)
3. Identify clusters — accounts that resolve to the same entity or share significant overlap are flagged as potential bot networks
4. Score and action — each cluster gets a risk score based on the number of linked accounts, the types of shared attributes, and behavioral similarity

The key advantage: this approach catches bots that individually pass all the standard checks but collectively reveal their artificiality through shared identity attributes.

Real-Time Prevention vs. Batch Detection

Most bot detection is reactive — batch analysis runs periodically and identifies accounts for suspension. But by the time the analysis completes, the damage (spam, manipulation, fraud) is already done.

Real-time entity resolution enables prevention at sign-up. When a new account is created, it’s immediately resolved against all existing accounts. If it shares a phone number with 50 other accounts, or a device ID with known bots, the system flags it before the account becomes active.

At Tilores, we process these resolution queries in under 10 milliseconds — fast enough to block a fraudulent sign-up before the user even sees the confirmation page.

Beyond Social Media

The Twitter bot problem is just a high-profile example of a universal challenge. Every platform that deals with user identities faces the same issue:

• E-commerce: Detecting fake review networks and promo abuse rings
• Financial services: Identifying linked accounts for fraud and AML compliance
• Gaming: Detecting multi-account abuse and ban evasion
• Insurance: Finding related claims that suggest organized fraud

In every case, the underlying technology is the same: resolve which accounts belong to the same real-world entity, find the connections between entities, and act on the patterns.

The Scale Challenge

The reason platforms like Twitter struggle with this isn’t a lack of awareness — it’s scale. Performing entity resolution across hundreds of millions of accounts, in real-time, while handling millions of new sign-ups per day, is an infrastructure challenge that few organizations can solve in-house.

This is why we built Tilores as a serverless API. The infrastructure scales automatically, the matching engine handles billions of records, and the resolution latency stays under 10ms regardless of data volume.

---

Building a platform with user accounts? See how Tilores can help detect and prevent fraudulent sign-ups in real-time.