How does Botometer detect bots?

Botometer detects bots by analyzing over 1200 features of Twitter accounts to discriminate between human and automated activities. This system utilizes a variety of approaches to generate scores for accounts:

1. Content Analysis: Botometer reviews the content, such as part-of-speech tags, based on linguistic properties, primarily for English language tweets.

2. Network Features: It evaluates networks associated with the account, like follower, mentioning, and retweet networks.

3. Account Characteristics: Features like account creation date, username characteristics (e.g., long names or presence of many digits), and profile images (like default or absent images) are considered.

4. Temporal Patterns: The frequency and regularity of tweets are scrutinized. Accounts tweeting at regular intervals or excessively are flagged, as human communications tend to be more irregular and bursty.

5. Bayesian Analysis: Botometer uses a Bayesian framework to statistically estimate the likelihood of an account being automated, taking into account prior knowledge about the proportion of bot and human accounts.

These features together help in making a determination about whether a given Twitter account is likely controlled by a bot or a human. Botometer is accessible via a website where users can input a Twitter handle to get scores, or through an API for more programmable access 1.