Hong Kong, the World and Facebook Trends
Will your news trend on Facebook?
Massive pro-democracy protests took place in Hong Kong last week. More than half a million individuals (most of whom were students) decided to occupy Central, the heart of Hong Kong. They prepared for a long occupation. Cleaning teams, supply and first aid stations were set up to enable an organized demonstration. As the crowd swelled, police used pepper spray, tear gas and batons. The reality was unnatural for most residents of this peaceful port, who began sharing photos and updates about what’s happening on the ground via social media.
For peaceful protesters, a potent offensive available is Media. Media spreads news of the protests, raises awareness and gathers support for the cause. However, we all know the predicament with media freedom in mainland China. Weibo, China’s version of Twitter, suffered its most massive censorship till date during the protests. Hours after Occupy Central started to take shape, Instagram was blocked in China.
Fortunately, some photos/videos of the protests had already escaped into Facebook and was being actively shared. One highly shared video (26.4K shares/9.5K likes) posted by Nero Chan showed the protests swelling as seen by a drone.
Media and Facebook
In addition to reporting the news, media can also spread, gather and maintain attention of the masses who aren’t physically close to the event. But how much attention would the rest of the world pay to Hong Kong? And how would we measure it, track its adoption and continuous maintenance?
Earlier this year, Facebook announced the launch of trending topics on its newsfeed page. Like Twitter trends, which reflect the attention landscape in the Twittersphere, Facebook trends showcase the most popular news stories within the Facebook world — algorithmically determined from publicly shared posts of its 757 million daily users. You can see a list of the top-10 most popular news stories on your Facebook newsfeed page.
I chronologically tracked five news events reported from Hong Kong during the protests and studied their evolution within the Facebook media ecology. Using Facebook trending data, I was able to spot which news stories became trends while which ones failed to, and which trends persisted while others died off.
My over-arching question revolved around the governing factors which decide if a news story will sustain as a trend in Facebook.
Trends sustain when enough people see a post and themselves share it. I found THREE distinct features that govern whether a news story sustains as a Facebook trend. While two of these factors were intuitive, one seemed unusual. Our data provides evidence that these factors led to disproportional attention regarding Hong Kong protests in different geographical regions. In fact, news stories about Hong Kong made it into the top-3 trends on Facebook only in certain geographical communities.
Facebook Trends Data
A Facebook trend is essentially a news event, which may be linked to several links/posts featuring the news story. Every trend comes with a headline (explaining the trend), related words (these are Named Entities and exists as Facebook openGraph nodes), a rank (indicating its popularity in the Facebook world) and the geographical zone the data was sampled from. Facebook currently provides trends across five zones: USA, Canada, Great Britain, India and Australia. Trends for each location are ranked from 0–49 in decreasing order of trending nature/popularity.
Online news media has rapidly transformed into a mobile, real-time phenomenon. Notably, Twitter trends was a powerful evolution in the domain of breaking news. However, the nature of Facebook trends is somewhat different from Twitter. Unlike the microblogging site, each trend in Facebook is not necessarily a multi-word or a hashtag. Instead, a Facebook trend is an event headline with related media (links). Facebook uses several natural language processing (NLP) algorithms that automate the task of attaching related media, topic extraction, summarization and headline generation for a link. Parsing natural language is quick, but not always realtime.
Thus, Facebook trends are slower to surface than Twitter. On the other hand, Facebook trends are richer in interpretability than Twitter because of included topic summarization, headlines and openGraph named entities.
Lets first look at how each geographical zone got initially exposed to the news. I’ll then explain how the factors led to this unusual adoption path.
‘Hong Kong’ News Diffusion on Facebook
I first plotted a persistence chart for the ‘Hong Kong’ trend on Facebook across the 5 geographical zones. A dot indicates the trend occurred in the geographical location at some particular time. Gaps indicate it fell out of the top 10 trending list.
Observe that the trend originated in Australia. It then started trending in India approximately 8 hours later, followed by Canada, GB in quick succession and finally made its way to the US.
What drives this special geographical route of acceptance, continuation and departure from attention on Facebook? To answer that question, we must comprehend what the driving factors behind trend-making on Facebook are.
Driving Factors in trend-making
To find the driving factors that make news stories into trends, we sample Facebook data every 5 minutes. Facebook provides us with at most 50 news stories that are trending at some time. On the newsfeed page, users can see only 10 of the most popular news stories (not all 50). What happens to the other 40 or so stories? These swim below the surface, competing with each other, trying to break into the visible top-10 list.
I track five events that unfolded over the first two days: (A) Protesters clash with police, (B) Thousands of activists occupy Hong Kong financial center (C) Police fire tear gas, (D) Police reduce force after 47 people are injured, and (E) Protesters begin stockpiling supplies. I found three key factors that influences attention on a news story and significantly decides it trending fate.
Luckily for us, the impact of all these factors is quantifiable. For my analysis and visualizations, I re-scale the ranking list from 0:49 to 9:-40. In other words, the highest ranked news story (most popular one) now has a score =9. The 10th most popular news story (and the last one to make the trend list/ visible to users) has a score =0.
(1) Time of Day
The time of the day when the story breaks is important. People don’t share when they are sleeping (at least we hope not). Diurnal patterns are common in social media, and there is no exception in Facebook. A piece of news that breaks late in the evening has a lesser chance of sustaining as a trend. There remains a possibility such a news piece might be picked up in the next morning though.
(2) Competing News Stories
Second, the number of competing news trends in a geographical community affects the trend sustainability. Competing stories reflects the ecological conflict that a piece of news faces to break into the top-10 list and maintain its spot.
Based on the number of competing news stories in a zone, we can calculate the Likelihood of trends making it into the the top-10 list. The lower the likelihood in a region, the higher the chances of the news story sustaining itself as a top-10 trend.
The likelihood of a news story entering the top-10 list is remarkably varied for different geographical zones. It is highest for Australia at 0.71 on average, followed by India (avg. 0.51), Canada (avg. 0.48), GB (avg. 0.46) and USA at (avg. 0.40). Its interesting to notice that the lower the likelihood value, the more choppy the trend line appears to be.
Notice that this sequence of decreasing likelihood is identical to the geographical progression of the ‘Hong Kong’ news diffusion.
(3) The Escape Velocity
Finally, how high the story rises in the trending list after it breaks and how long it lasts there plays a crucial role in the story’s eventual longevity.
Lets track news about events in Hong Kong that break almost at the same time in two geographical locations. For the same story at breaking points [E] and [F] in Australia and India respectively, notice the former trend sustains while the latter does not. For another story at breaking point [G] and [H], the former trend which was among the top-3 sustains whereas the latter fails to sustain in the next 2 hours. The trend in Australia reaches a top-3 slot in both occasions, and ends up sustaining for ~16 hours. The trend in India falls of the top-10 list quickly after it breaks.
I observe similar phenomenon in other geolocations. Consider the same news story (Police firing tear gas) which breaks at the same time in Canada and India as [D] and [C] respectively (shown below). The trend in Canada starts in the top-3, and continues to trend for the next 23.5 hours. In India the trend starts as the 7th-most popular. It drops out of the trending list after just 1.5 hours.
The marked influence of top-3 trending slots is not just specific to news about Hong Kong. It is evident in other news stories that break throughout the day. For example on Aug. 29th, 143 news stories where reported on Facebook in the US. Only 61 of these stories made it into the top-10 trend list (visible). 27 of those 61 were able to reach a top-3 rank at least once. And 17 news stories out of the 27 that reached the top-3 lasted there for at least 1.5 hours.
Out of all the available trending slots in a day, 70.3% of slots were occupied by trends that had been among the top-3 for at least an hour and half. In other words, those 17 news trends occupied almost 3/4th of the available trending slots in that day, competing against 126 other news stories.
The influence gets more drastic in other zones. In Canada, 40 news stories out of an overall 90 made it into the top-10 trending list. But only 15 out of the 40 stories lasted as top-3 trend for an hour and half. These 15 news trends occupied 74.4% of the trending slots. In India, just 10 out of overall 78 news stories which reach the top-3 end up occupying 73.6% of the trending slots.
In fact, between Aug. 26th and Sept. 4th, I found that only 12% of news stories which break on a day end up occupying ~72% of the trending slots for the next ~16–18 hours. What’s common among these 12% of stories? They all had risen as high as the top-3 trend and had survived there for at least 1.5 hrs. In fact, stories that did not last for 1.5 hrs in the top-3 had a 57% chance of falling off from the top-10 list within the next 6 hours. Thus, the first 1.5 hours in the trending list is critical in a news story’s longevity and a powerful symptom for trend sustenance on Facebook.
What’s special at the top (3)?
So what’s special about the top-3 territory? Here’s a potential reason:
On your Facebook newsfeed page, you will initially always see only 3 trends — the top 3. You need to further click ‘see more’ to view all of the top-10 trends.
In other words, the top-3 trends are immediately visible to every user who loads the newsfeed page. The next 7 trends (marked in red) are seen only by users who explicitly click the ‘see more’ button. We can assume the latter group have their mind set on exploring popular news being shared in the Facebook world.
Its incredible that a subtle design choice can have such an enormous impact on how news trends sustain in the Facebook trending list. One might wonder what categories of news usually lasts in the top-3 slots? Our data shows these are overwhelmingly stories related to politics, sports and entertainment.
Data paints an interesting picture of the Hong Kong protests, how it spread via Facebook, where attention was maintained and why. But, what’s actionable from all this information? What can we conclude about trending patterns on Facebook?
(1) First, it is important to remember that a combination of all three aforementioned factors contribute to which news trends and keeps trending on Facebook. (2) But, the time of the day when a story breaks on Facebook (or is shared for the first time) plays a bigger role in more competitive news ecologies, e.g., USA and UK compared to Australia. (3)Finally, the initial number of shares within the first 1.5 hrs since the story breaks is critical in giving the news enough ‘escape velocity’. This velocity enables the trend to last in the top-3 long enough.
Maintaining a top-3 position allows not just the ‘news-philic’ but casual Facebook surfers to stumble upon the story. This leads to more shares and the story reaching a critical mass of users. Increased sharing sustains the trending position. And that feedback loop of trending is complete.