Driving Trending Topics Home

The local influence in what becomes global news

The tweeting population of a city is a intriguing group. Its hard to pin point what intrinsically drives them to tweet, but what matters is when they collectively tweet about some topic — it shows up on Twitter’s trending topic list (TTL). For many of us, an easy way to be informed about ‘whats happening in the world’ — is by tracking the Twitter trending topics.

The TTL of a city (e.g., New York) is considered to be reflective of topics that have captured the city’s attention (here is how Twitter’s algorithm calculates what’s trending). The assumption is that sustained trending nature of some topic is a proxy for ‘sustained attention’ on that topic.

But more often than not, questions arise around why an event happening in the real world — is NOT trending on Twitter.

More on the geographical dispersion of #FreddieGray below, and why it did not trend nationally.

There are the usual suspects as to why something isn’t trending on Twitter: not enough media coverage, lack of consolidated hashtag usage, competing news stories or perhaps algorithmic censorship. But these conjectures is just the tip of the iceberg when it comes to geographical trend dispersion on Twitter.

As you can see, people are not happy about things ‘not trending’ on Twitter.
Recently, we published a paper (preprint here) called ‘The Attention Automaton’. In it we show how to model, to a significant extent, the Twitter trending ecosystem with respect to different geo-locations and specific categories of trends.
We found that geographical trend dispersion is primarily determined by three factors: the information category of the topic, the attention shift tendency of the location and the initial persistence of the trend.

Given sufficient information about these factors, it could be possible to identify ‘The Perfect Seed’: a single (home) location for the trending topic where it must persist initially in order gain momentum and in time trend everywhere, eventually exploding into national attention.

The Method

We represent a city’s attention to some topic by the duration for which it lasts in the TTL for the city. We collect TTL data for approximately 121 world locations at intervals of 5 minutes. The dataset has been populating since 2012. Using a computational model called probabilistic automaton, we were able to simulate the dynamic changes in Twitter geographical TTLs, essentially predicting which trends will remain in the TTL while which ones will drop off over time.

The study found that cities were disproportionately receptive to specific categories of trends. The affinity of a city for a trend category is determined by how long it keeps trending topics of that category and how often it initiates such trends. For example, San Francisco has strong affinity towards trends in Gaming and Technology, whereas Boston has strong affinity to trends in Politics and News. Most business trends originate in Chicago or New York. The most sensitive city to news is ..wait for it…. Baton Rouge. A large number of lifestyle trends originate from Los Angeles and.. Chicago. South American cities have extremely high sensitivity to music and memes, but extremely low sensitivity to news, politics or business trends.

The Volatility SNR (stands for Signal-to-Noise Ratio) of the city is a statistical measure which indicates its attention shift tendency. The higher this value (and size of bubble), the greater is the attention shift tendency of the city, which makes it more prone to catch new trends. On the other hand, it is also difficult for a trend to persist in these locations if it does not belong to one of the affinity categories.

The study also revealed that cities have wide variation in how quickly the attention shifts between topics (based on how quickly the TTL changes). The rate of TTL change in the city is measured by a term we call ‘Volatility’ and indicates its attention shift tendency. Volatility regulates a city’s potential to sustain a trend. Faster the TTL changes, higher is the volatility and the attention shift tendency. The city with the fastest attention shift is — Tokyo. New York comes second, Washington DC third, Los Angeles fourth and London fifth. Higher attention shift tendency means greater chances of catching a trend, but it is also harder to sustain the trend if it is not in one of the affinity categories.

Based on these factors, let’s get back to the question of why some hashtags trend nationally and others do not.

(1) #FreddieGray

Freddie Gray, a Baltimore man injured during an arrest by Baltimore police last week, died Sunday at Shock Trauma on April 19th, prompting protests by city residents and activists. Promises were made by the city officials for a thorough investigation. #FreddieGray was the hashtag most people used to raise awareness about the incident on Twitter.

As Gilad mentions in his post, there was lack of consistent trending (or persistence) in the #FreddieGray hashtag on Twitter. It could be one of the reasons the trend did not reach all the big metropolitan cities or go national.

In fact, if you look at the difference in time span between the trend spreading to two consecutive cities, the lack of steady dispersion becomes immediately visible. Compare this with the Ferguson trend drift and we notice the sheer difference — none of the known news hubs (cities receptive to news) produced a strong initial persistence of the hashtag #FreddieGray.

The bars indicate the amount of time spent until the next infection of a city by the trend. Infection here means spreading to a newer city/ the trend appears in the TTL of a new city

Our data shows that Minneapolis, Atlanta and San Diego do not have strong affinity for news stories. Both Minneapolis and Atlanta both heavily receptive to sports while San Diego has affinity for technology trends. Thus, even if a news trend (e.g., #FreddieGray) appeared in these locations in the initial phase, the affinity mismatch would hamper the initial persistence of the trend. Consequently, #FreddieGray fails the initial persistence condition and doesn’t spread to other cities which are more receptive to news (New York, Baton Rouge, Chicago, Boston etc.). The trend persists in Baltimore for 22 hours without spreading to a single new city on the day of the event.

And although #FreddieGray did trend discontinuously in some US locations for a short periods between April 25–27, something else was pillaging the focus:

Alternately, let’s analyze two worldwide trends that captured global attention. (I have previously written about Ferguson and Hong Kong protests in-depth, so I will skip those events).

(2) #mh370

Consider the Twitter discussion around flight MH370 that disappeared on March 8, 2014. Here is an interactive view of the #mh370 trend spread worldwide and through the US.

The #mh370 trend on Twitter lasted more than 3 days. After commencing in Klang, Malaysia on March 8th, 2014 at 1:27 PM Eastern time, it spread to the US within 19 mins of origin, captivating the entire nation (70% of US locations) within the next half hour!

It took 6.8 hours since origin to reach India (probably around the time everybody was waking up there and heard of the news). Surprisingly though, it appeared in the UK after approx. 2 days after the story first broke. The trend spread to mainland Europe shortly after.

Looking specifically at the US, the #mh370 trend had an interesting drift. The first US location to grasp it was Baton Rouge. This is a pattern we have observed with regard to other ‘news’ trends too — Baton Rouge is extremely sensitive to news (low volatility, high persistence in news). Observe how delayed Los Angeles is in trending #mh370 compared to many other US cities. Cities have an inherent affinity to certain categories of trends, news or otherwise. Los Angeles is extremely sensitive to trends about lifestyle and entertainment, but not breaking news.

Jakarta is the 117th country on #mh370 trend drift chart. This is almost 3 days after the event happened involving its neighboring country. For trends around One Direction (+music), Jakarta’s sensitivity is almost instant. Category of the trend plays a huge role in the city’s receptiveness to it.

(3) #jesuischarlie

Lets look at the #jesuischarlie hashtag around Charlie Hebdo shootings. Here I plot the sensitivity of cities in catching up to the trend (x-axis) given their geographical distance from the event location (y-axis).

Within the first 60 minutes, #jesuischarlie trended in most French cities (expected) and some British cities but not London. In the second hour, it spread to cities in Spain, USA and Germany. The first Canadian city to pay attention is Montreal, far ahead of Toronto or Vancouver. This can be attributed to parts of French Canada being more sensitive to news about France. There is sufficient initial persistence (within the first 3 hours) to sustain the trend.

In Spain, almost 4 hours elapse between Madrid and Barcelona trending the story, in spite a high cross-network connection among individuals in these cities. This is because Madrid has a much higher attention shift tendency than Barcelona. Similarly in Australia, Adelaide and Perth is more attentive to the story compared to Sydney! Again, Sydney we found had a lower attention shift tendency than Adelaide or Perth. In Stockholm, #jesuischarlie appeared within 6 hours of origin (but #mh370 never trended there at all).

Stockholm, Milan and Barcelona all have low attention shift tendency (low volatility). Lower attention shift means less chances of catching a trend. We found that London and Stockholm both have low affinity towards ‘activism/cause’ category of trends. In this category, the city with the highest affinity if New York (one of the critical drivers of #Ferguson trend). As a result, notice that New York is one of the first American cities to catch #jesuischarlie.

At this point and with regard to these two worldwide trends, I must say that there is possibly other finer factors at play within the location’s population — their anxiety about terrorism, their admiration of journalism and free speech and even their fascination with conspiracy theories and trolling.

The Perfect Seed

Which brings us to the obvious next question:

If we knew the category of the news topic and the attention shift (volatility) in all these cities, would it be possible to find which city is optimal to germinate the trend, i.e., get the topic into the TTL of that city first, so it can spread to other cities and develop into a national trend.
We must choose a germination location that has (1) affinity to the category of the trend topic, (2) significant volatility to catch the trend, and (3) where initial persistence can be forced.

We do not have an ultimate equation for this — YET. What we do have is three critical parameters — the category, the attention shift tendency and the initial persistence. Using permutations of these parameters, we did find some interesting patterns within geographical communities and their receptiveness to trends that explode into national visibility.

The results culminated into an interesting hypothesis: if you want to make something trend in a super-volatile noisy city like New York, perhaps you should carefully inspect the trend category and find a city where it can persist in the TTL for a critical period to gather enough momentum. A trend is like a ripple moving in a sea of undulatory attention, where the dynamics of the sea is as important as the characteristics of the ripple generator.

And then there is the issue of Twitter bots, and what might happen when they figure out the possibility of an optimal seed location! Perhaps they will begin to embrace proxy addresses from an American city and not some Ukrainian village when they tweet. But until we find that home city where a trend can germinate, grow and gather enough momentum to eventually attain national focus, we should continue to use, contribute, analyze (and despair) at Twitter’s trending topics.

Questions/Comments: Find me at @_RoySD or suman@betaworks.com. Thanks to Gilad for helping with the draft of this post.