Feature extraction identifies and derives attributes that represent relevant characteristics of data points or graph nodes. In a user graph, extracted features become node properties that describe behavior and support later comparison. section 8.0 section 8.1
A practical user-segmentation workflow extracts behavioral features such as:
- tweetCount: the number of tweets published by each user, including users with zero tweets. section 8.1
- retweetRatio: retweets divided by the sum of retweets and tweets, using zero when both counts are absent. section 8.1
- timeToRetweet: the average duration in minutes between publishing a tweet and retweeting it, using 372 minutes for users who never retweeted. section 8.1
Measuring retweet delay
A Cypher query matches users, retweets, and original tweets; calculates duration.between(tweet.createdAt, retweet.createdAt); averages each user's delays; and returns distribution statistics. Very short, repeated delays can help identify likely bots, while the observed average for ordinary users is around six hours. section 8.1
After extraction, feature vectors can be compared with pairwise cosine similarity to create a network whose links connect similar users. Community detection can then reveal user segments without predefining their number; network density influences cluster size. section 8.0
