Social Bearing recently launched a new feature where you can sort twitter users by a new metric; the ‘TUQI Score’ or Twitter User Quality Index. This is applied to user entities within user search, followers search, friends search and people search.
The TUQI score is an attempt to rate the quality of Twitter users by various metrics such as friend/follower ratio, tweet/follower ratio, last tweet date, profile completeness and other factors which are all available through the Twitter API.
Twitter users with high TUQI scores are more likely to be interesting, active and engaging whereas Twitter users with low scores are more likely to be a sign of spam accounts or less active and engaged users.
The TUQI score can help people judge who to follow or unfollow as well as identifying spammier looking accounts, or accounts that may use automated follower techniques which can inflate follower counts.
There are many measurements contained within each Twitter user entity through the Twitter API which can be combined to help us give a good idea of which user accounts are higher quality than others.
So what factors can we glean from Twitter API user data which can be used to help determine quality?
Follower count on the surface may seem like one of the best candidates to rate quality Twitter users. It goes without saying that people generally acquire followers because other Twitter users like what they tweet about. Someone with a high follower count therefore means that a lot of people find them interesting.
The problem with using follower count alone to rate quality is that it will often yield poor results. There are a number of methods for example that can inflate a users followers count, often based on en-masse or automated follow-back tactics; i.e. I follow you, you follow me (and if you don’t, I will unfollow you).
I also frequently come across users who have over 10,000 or 100,000 followers yet they follow a similar amount of people themselves. Rather than manually following this amount of people, it is much more likely that automated tools have been used to acquire followers. These automated tools will find and follow 1,000’s of people on behalf of a Twitter user and if they are not followed back, then the tool will unfollow them again.
By not using Follower count in the TUQI score means I don’t have to account for this inflated follower factor.
A persons follower count compared to the number of people they follow is a good measure of an interesting or integral Twitter user.
Let’s say person A and person B both have 10,000 followers but person A is following 15,000 people whereas B follows 1,000 people. This would give respective friend/follower ratios of 1.5/1 and 1/10. It is much more likely that person B’s followers have been acquired more naturally as a result of people liking what they tweet about, rather than (for example) their followers being acquired as a result of users being followed themselves, as more likely with person A.
The friend/follower ratio therefore plays a big part in calculating the TUQI score
The amount people tweet, could potentially be used in the quality rating system. People who tweet infrequently are less likely to get retweets, shares, engagements and followers. Conversely, someone who tweets too much may annoy some of their followers.
Tweet totals however are not really a good indicator of how interesting a Twitter user is. Other factors such tweet/follower ratio and time since the last tweet are better measurements.
The more your tweets are seen and shared, the higher the chance you will gain more followers, but tweet volume is a poor indicator of quality. If user A tweets on average 20 times a day, gaining 2 followers a day and user B tweets once a day but gains 1 follower a day then this gives respective tweet/follower ratio of 20/1 and 1/1. Although user A has more tweets and more followers, user B is gaining more followers per tweet which would be a better indicator of having more interesting tweets.
The tweet/follower ratio plays a strong role in the TUQI score
The number of total friends can play quite a large part in calculating Twitter handle interest scores. Someone who has followed 100,000 people for example is unnatural behavior because they are:
– Unlikely to engage with their friends
– Unlikely to read the tweets of 100,000 people (2,000 people would be hard to keep up with)
– Likely to be following this amount of people for the main reason of getting follows back
– Likely (but not always) to be using automated follow/unfollow tools.
Friend count in the TUQI score starts to take effect after 5,000 friends
People usually do not want to follow inactive or very infrequent tweeters. Someone who may have once been active and engaged on Twitter will be far less interesting in the eyes of their followers if they haven’t tweeted for months.
The date of last tweet has a varying effect on the TUQI score and begins to take effect after 7 days. A user who hasn’t tweeted for 100 days will have a much lower TUQI score (all other things being equal) than someone who hasn’t tweeted for 10 days.
Being a verified user may be an indicator of an interesting and exceptional person in the world but does not necessarily mean they are an interesting person to follow. Verified users will usually acquire significant numbers of followers by virtue of who they are outside the Twittersphere, rather than the quality of his or her tweets.
Verified users will have a strong inverse effect on the TUQI score as a sort of leveling effect. Otherwise verified users will almost always be at the top of any interest based calculation as they will have disproportionately high tweet/follower and friend/follower ratios.
The number of lists a user belongs to can be a good indicator of how genuine and interesting other Twitter users have found them and taken the time to add them to curated lists.
Although most people probably don’t use Twitter lists, the list count can still be a good indicator of natural and engaged Twitter users
The list count has a small positive effect on the TUQI score.
It would be great to be able to use the total number of retweets and favorites of a user as a proportion of their followers in order to help determine quality. This unfortunately isn’t possible as total retweet or favorite count isn’t in the user data returned by the Twitter API. In order to get retweet totals for example, we would need to loop through a decent sample size of each individual user’s tweets. This would take time and quickly hit Twitter API rate limits.
The number of times a user has favorited other tweets however is returned in the user data. This can be a good indicator of how genuine and engaged someone is and they are more likely to be reading their friends tweets.
Fovorited count plays a small part in calculating quality scores.
A decision to follow someone or not can be often be driven by what interests, hobbies and other information people put in their bio. If the bio description is empty, a user is less likely to be followed. Bio completeness therefore has a small impact on TUQI scoring.
A lack of a profile picture often indicates accounts that have been setup for spam purposes, or new users taking a look at Twitter for the first time. The vast majority of active, natural and engaged twitter users will have an avatar of some kind rather than the default egg and so lack of a profile picture will have a negative impact on TUQI ratings.
Reviewing users tweets before deciding to follow someone is important to many people. Protected user accounts prevent this and so reduce the TUQI score