Some people asked where I got my information from for the article “Six Secrets About Whether Twitter Censors Trending Topics” at the Buffer Twitter Tools and Twitter Management blog. I have written this post to address that. (And if you haven’t, you should read that first.)
99% of the “controversy” about Trending Topics that I’ve seen is of two types:
- They think it didn’t trend, but it did—they simply missed it when it did;
- They don’t understand why popular topics don’t keep trending. If they did, Trends would be mostly things like “love,” “hate,” “Justin Bieber,” etc. Trending is about more than just simple popularity.
(In other words, most “controversies” are based on misinformation.)
I’ve relied on three sources of information for determining how Trending Topics are calculated. By combining information from these three sources, I’ve come to certain conclusions. Some conclusions are clearly true; some are only possibly true.
I’ve then taken the conclusions and looked at the information again, to see if it helps gain further insight into the source information.
My three sources for information about Trending topics are:
1. Info from Twitter explaining how trending topics are calculated.
Some of what they say is clear, some of what they say makes certain conclusions likely, and some is unclear.
To Trend or Not to Trend… Key quotes:
- Sometimes a topic doesn’t break into the Trends list because its popularity isn’t as widespread as people believe.
- Sometimes, popular terms don’t make the Trends list because the velocity of conversation isn’t increasing quickly enough, relative to the baseline level of conversation happening on an average day.
- Topics break into the Trends list when the volume of Tweets about that topic at a given moment dramatically increases.
- Twitter Trends are automatically generated by an algorithm that attempts to identify topics that are being talked about more right now than they were previously.
- The Trends list captures the hottest emerging topics, not just what’s most popular.
About Trending Topics. Key quotes:
- The following behaviors and others like them could cause your account to be filtered from search or even suspended…Repeatedly Tweeting the same topic/hashtag without adding value to the conversation in an attempt to get the topic trending/trending higher.
- The most important thing is to make sure your Tweets are genuine thoughts or impressions and not attempts to insert yourself into a trend. Everyone who clicks on the trending topics should be able to see real people’s ideas and links to further relevant information.
Tweets from Twitter management about Trends (e.g @DickC). Key quotes:
- …we don’t block topics from trending, we only remove a few specific obscene terms
- …trends are algorithmic, not chosen by us but we edit out any w/ obscenities & I’d like to see clearly offensive out too
An infographic Twitter recommended about Trending Topics. Key quotes:
- Twitter Trends favor novelty over popularity.
- The…algorithm only accounts for interesting peaks: sudden increases that mark an emerging trend.
- Twitter used to rank popularity by volume, but changed the algorithm.
- …the Bieber effect; becoming part of the constant background noise like love, hate, Christmas [etc.]
The first three of these sources were linked to in Six Secrets About Whether Twitter Censors Trending Topics, although the links weren’t particularly obvious.
2. Investigating how the volume of tweets affects topics that have trended
Because it’s hard to know what localization data Twitter is using, I’ve only paid much attention to observations about volume vs. trending for topics, that:
- …would clearly be popular primarily in the U.S., e.g. use slang primarily popular in the U.S.
- …trend worldwide, so location data is irrelevant.
3. Observations of the differences in Trends when the algorithm was first implemented.
What do we know about Trends, and what can we figure out?
There are potentially several factors that might affect whether a topic trends:
- Are there more tweets about it than about other topics that are trending currently?
- Has it trended before?
- In what geographic area are its tweets coming from?
- How much of the tweet volume is from a variety of people, and how much from the same people tweeting the topic repeatedly?
I think some of the key questions are:
- Why does a topic of sufficient tweet volume not trend for the first time?
- Why does a topic of sufficient tweet volume not continue trending?
- Why does a topic of sufficient tweet volume not trend a second time?
- Does Twitter minimize “the same people saying the same things” as counting towards trending?
(“Of sufficient tweet volume” means that a topic has a tweet volume in a geographic area that is higher than the topic with the least tweet volume in a geographic area currently trending, e.g. if volume were the only consideration, it would appear to be worthy of trending—though you would have to check other competing, non-trending topics to be sure.)
What conclusions can be drawn?
Twitter does say that trends are affected by the “volume of tweets…dramatically increas[ing]“. So sheer volume is not the only factor. A gradual increase in tweet volume about a topic to a level of “sufficient tweet volume” to trend might then not trend. So that can explain question #1: The topic is not a novelty enough; is not “dramatically increasing” enough.
Question #2 can simply be because at some point, a topic must become “old news.” So maintaining “sufficient tweet volume” won’t keep an item trending indefinitely.
Question #3 would seem to be easily explained as “because it’s not a novelty anymore” and yet, significantly, topics have trended a second time in the same area (point #1 below). So this leads us to search for a possible hidden factor.
The most important thing that Twitter does NOT make clear is the answer to question #4. However, my conclusion is that Twitter does minimize people in the same area saying the same things towards contributing to a topics trending again or continuing to trend. Or, more accurately, they weight people in new areas joining the conversation more highly than people in the area where tweet volume originally caused a trend continuing to tweet. Here’s why, with each point leading to the next:
- Twitter says that the same person “Repeatedly Tweeting the same topic/hashtag … in an attempt to get the topic trending/trending higher” may be filtered out from counting toward the topic. This indicates there is at least one mechanism for counting volume but eliminating some people. But more importantly, it shows that there is an inherent need for a minimum number of people to be tweeting. For example, a small number of people each “overtweeting” a topic to the point that total tweets could reach a trending volume could ALL be excluded from being counted due to overtweeting.
- I have seen topics that trended a second time in the same area, so it is not impossible. Of particular note is that this means there is likely something in addition to volume that is not novelty that can cause a second trend. I think that something is new people tweeting about the topic (in addition to sufficient volume).
- When the algorithm was first introduced, Justin Bieber fans made an enormous effort to create a volume of tweets higher than anything they had previously achieved when they saw that their usual efforts didn’t cause trending. Despite having a very organized network that was repeatedly successful in creating high tweet volumes of his name before, they were unsuccessful afterwards. It can’t only be “lack of novelty” that caused the topic to trend, since it is possible. It appears to have been “the same people saying the same thing” not being counted highly.
- Twitter says, in the same article, “Topics break into the Trends list when the volume of Tweets about that topic at a given moment dramatically increases.” but also says “Sometimes a topic doesn’t break into the Trends list because its popularity isn’t as widespread as people believe.” The first quote indicates that “volume” is required, the second that “widespread” popularity is required. The implication is that “widespread” does not mean the same thing as “volume.” Volume would be a count of tweets, but “widespreadness” would be a count mainly of people, or variations in geographic area with the same trending area. This is admittedly not definitive, but considering in particular point #2 above, this seems to be the case.
Does Twitter Manually Censor Non-Offensive Trends?
When something looks non-random, in the absence of other explanations, people often suspect there is some kind of force or interference at work. When you see a lot of tweets about something and it isn’t trending, it’s tempting to think that it’s been blocked from trending. But first you have to understand what trends are, and then you have to do some basic research, such as seeing if the topic did trend, but you missed it.
Of course, anything that trends gets more attention put on it, more people involved, etc. But for topics that have a high volume of tweets without officially trending, clearly just being listed as a trend is not a key issue in making the topic popular. That makes the distinction of being a trend pretty minor. To take the example to the extreme, let’s say 100% of all tweets around the world are about a particular topic, and it isn’t trending. In that extreme case, there is no benefit to it trending, since that’s all anyone is talking about anyway.
Although that’s an extreme case, most cases people complain about are similar. Things that are being widely tweeted about…are being widely tweeted about. They have already achieved popularity. Very little would be gained or lost by Twitter manually censoring them from trending.
Also, Twitter has resisted providing many different governments with information on Twitter users. Yet some people in virtually every country believe that Twitter has censored political trends in their country on the request of their government to Twitter, or due to bias on Twitter’s part.
Twitter has demonstrated that they are the most free-speech, anti-government interference large internet company that has ever existed, yet some people still believe they censor political topics from trending, even though to do so would have very little impact.
I’ve put how I consider Trending topics to work into the aforementioned article “Six Secrets About Whether Twitter Censors Trending Topics,” so I won’t repeat that here.
Why Twitter needs to count more than popularity
Some people say they wish volume counted for more; it bothers them that something that is very popular isn’t trending. But if they got their wish, they would probably still find their chosen topic doesn’t trend.
Why? Because if anything popular trended, things like “love” and “hate,” etc. could be trending.
Are Twitter’s Trending Topics Broken?
I would say yes, because they confuse almost everyone at some point or other. Twitter could provide an official “data” page, which would include, among other things, the volume of perhaps the top 50 items per trending topic area, and allow people to click on the ones they wish to filter out. Twitter could run ads all over such a site without too much objection, since people would be voluntarily visiting to check the data.
Is My Analysis Any Good?
What Twitter trends are trying to do is pretty simple, and it’s fairly easy to describe in general. I haven’t said anything particularly groundbreaking here, other than extrapolating a bit from pointing out that Twitter has made statements that they can count people and not just tweets, and has provided examples of their ability to do that.
I am open to the criticism that I have no data backing up my own investigations when I say what I personally have observed about Trending Topics, because each case I checked out on someone’s request was so obviously NOT a case of censoring I didn’t bother saving any information. Every time I have checked, it turned out the person bringing the situation to my attention did not understand the basics of what Twitter means by “trending” nor did any simple checks themselves. (Most commonly complaining about something that DID trend but they missed it.)
Cases I checked out for my own curiosity were also very clear cut, and at the time I never expected to be asked about them. I research dozens of things each day (mostly via Google) and like most people, I don’t keep any notes. I just dig in until I’m satisfied.
There are tools that can be used to go back and check trends vs. volume, and so if anyone has saved information, or wants to share research, I will more than happy to include or make mention of it here.
UPDATE: Great analysis of Occupy Wall Street trends across several hashtags by Gilad Lotan, VP of Research and Development at Socia.