Untangling the media

Reflections on the state of traditional, online, and social media, and on attempts to understand them

It is quite clear that the media has a large influence on how we perceive the world. Indeed, almost all our knowledge of reality comes not from first-hand experience, but other sources such as TV, newspapers, film and the internet. The massive increase in availability of such forms of information in the past century has made us much better informed, being able to get a reasonable idea of events happening on the other side of the planet in almost real-time - something which was completely inconceivable not too long ago.

As we consume more and more media, it becomes not just a way of finding out about what's happening elsewhere: it is reality. And since this reality is based on second hand information, fed to us by third parties with their own aims and agenda, we can easily get a false view of the world if the information is inaccurate or biased. This need not be deliberate deception, or even a conscious manipulation of facts, but a consequence of providing information in a way in which it is most easily absorbed. For example, the portrayal of violence on television (not just in news but also films and other fiction) is known to be far in excess of its actual incidence. This has various causes - sensationalism, the fact that violence shocks and sells copies, and because it's what viewers expect to hear - and people then genuinely believe their environment to be a more dangerous, unpleasant place than it actually is (this is called the mean world syndrome).  Similarly, people's views on scientific topics, such as climate change and the state of the economy, are often incorrect and grounded in misleading and biased reporting.

These issues were discussed at a lecture I attended a few weeks ago, hosted by the University of Bristol. Justin Lewis, a professor of communication at Cardiff University, discussed his work studying people's consumption of various media, and how this affects their beliefs about the world. In his work he has studied people's reaction to climate change, and how even though this is one of the most dangerous, urgent problems that humanity faces, we are in general remarkably relaxed about the whole thing.

It is interesting to note to that, while blatantly biased reporting obviously plays a role in all this, one of the ways in which people come to believe falsehoods is by erroneously imagining a connection between frequently co-occurring facts. For example, there is a reasonably widespread belief that the Earth is warming due to the hole in the ozone layer; this is totally untrue, but the point is that they are both often mentioned together, and this gives people the false impression that they must therefore be related, and even to construct an illusory causal relationship. Another example is the disturbingly pervasive belief in the US (and apparently in the UK) that Saddam Hussein had a role in the 9/11 terrorist attacks, in league with Bin Laden: this is nothing but paranoia, but it is very likely that the two figures were frequently mentioned in the same news items, leaving people to inadvertently come to their own specious conclusions. Such unintended interaction between unrelated pieces of information would be hard to predict and track, and harder still to counter; it is a property of an incredibly complex system of information, over which individual people or organisations have a diminishing amount of control.

An interesting point made by Lewis was that, in this age of unparalleled access to information, where almost the entire knowledge of the species is available at our fingertips, people will still tend to seek out what is familiar to them. So rather than find new perspectives, people tend to go for that which confirms what they already know, and fits in with their current world view. This is quite understandable, since we generally prefer that which is familiar; but it encourages media providers to provide what they know their public wants: they make sure people hear what they expect to hear, packaging stories in a way that is most appealing (this is clearly evident in newspapers such as the UK's Daily Mail, well known for giving a hysterical, fear-mongering, xenophobic slant to everything; as well as traditionally more 'upmarket' publications such as the Guardian, whose left-leaning stance on issues such as immigration and the environment can usually be expected to chime with their readers' expectations.) The unfortunate consequence of this is that people's view of reality comes from the media they choose to consume, which is produced by organisations who sell them what they want to hear - a self-perpetuating cycle of misinformation and distortion. This is rather worrying, not least because politics and the media are so often intertwined, with policies aimed at appeasing papers and their readers - but also because such feedback loops and cycles of obfuscation become increasingly hard to comprehend.

The attempt to understand the way media is generated and consumed is the research interest of the second speaker at the lecture, Nello Cristianini, who followed Lewis's talk with an overview of the work carried out by his research group here at Bristol. The vast output of all the world's media is far beyond the capability of any human to comprehend, but using software which they have developed, it is becoming possible to automatically analyse news reports, from a variety of outlets, in an attempt to understand the constant flow of information. Obtaining, processing and storing the data is a feat of engineering in itself, requiring some rather substantial computing resources; but the real challenge is to autonomously understand the content of these articles.

One interesting area of this work is analysing which news articles become popular (defined as those that make the 'most popular' list), and attempting to predict this. It seems that it is not possible to simply classify, for a given article, whether it will be popular or not (understandably, since what is popular changes from day to day); but given a set of news articles it is possible to quite accurately predict which will be more popular than others from the same day. One thing that becomes evident from such analysis is that stories about people and celebrities tends to be more popular; and even amongst the readers of outlets specialising in topics such as politics and business, the articles they tend to prefer are of a more tabloid variety. This is perhaps not too surprising, and fits with the idea mentioned above that people will tend to seek out what interests them the most. I can imagine this kind of research being used by content providers to more accurately target their consumers with more of what sells well; but also as a useful tool for understanding how people consume information, and the most effective way of disseminating news stories to people so that they remain interested.

However, one thing which is important to realise about news stories is that they are not, in fact, "stories". Humans are a storytelling species, and our voracious consumption of novels, film, soap operas, musicals, opera, comic books, and theatre go to show how engaging a well-told tale can be. However, as we discussed after the lecture (I was fortunate enough to have a drink and a chat with the two speakers and a few of their students), news media is not presented this way. There is a headline, which essentially states the main, salient point; there is a preamble outlining basically what happened; and then there is the main body of the article, outlining the finer detail and background to the event. Maybe it's this that drives people to be disinterested in weightier topics, and seek items which are more easy to relate to personal narratives, or have a more direct relevance to their own lives. It's interesting that news coverage is almost the only genre that does this: in film, music, computer games, there is always the tension caused by not quite knowing what will happen next. Newspapers did not do this; radio and television news followed this model, with short snappy headlines followed by a progressive spiral of illumination; online news outlets naturally fell into the same format.

I must admit though, I do have some reservations about this hypothesis, and can't quite imagine actual news presented in an exciting, narrative-driven fashion, with key facts being withheld until the gripping conclusion; news written in this way would, I suspect, be infuriating.

In this discussion of how news stories aren't stories at all, another obvious example came to mind that takes this hierarchical approach to disseminating information: academic papers. Here, even more than in news articles, there is a concise, informative title, followed by an abstract quickly summarising the purpose, methods, and conclusions of a piece of work, followed by a lengthier account of the technical details. As my supervisor in my earlier days used to tell me: this is not a thriller, don't withhold information - tell them what they need to know and give detail later. This may well be a more efficient means of describing research after all - though as Justin Lewis lamented, few people actually read these papers. A dry, no-suspense delivery style might have something to do with it (of course, the lack of attention span is exactly why we structure papers like this - we have abstracts because we know they don't read the whole thing).

But this pessimism about the homogeneity of individual's media consumption may not be so well deserved. After all, some stories/articles/videos become immensely popular, and spread over the whole internet with virtually no top-down control, exposing people to viewpoints they might not have actively sought - that is, they go viral. Naturally enough, our conversation turned to memes - where ideas are considered replicators, spreading through a population simply because they are good at spreading. Perhaps this could be a way of spreading information, in an easy to consume format: produce many variants of a news story, release them, and the ones which are better at spreading will reach the widest audience.  Exploiting the new-found knowledge of which kind of stories people want to read could make such an approach more likely to succeed (indeed, the secret to creating viral, self-perpetuating content has long been a goal of advertising and marketing). Whether packaging news in viral memes and spreading them - essentially tricking people into reading what an editor thinks they should be reading - is a good idea, I'm not so sure; then again, packaging content in an easy to access, relateable format is exactly what newspapers, TV and the internet have always done (I should emphasise that I'm not talking about memes in the lolcats / face-with-caption sense, but rather the construction of news stories in which makes people want to share them).

Of course, consumption of top-down media is only part of the story. User generated content, such as blogs and social networking, are becoming a significant factor in the way people react to and find out about the world; and as complicated as older media styles are, these promise to be another level again. But again, recent research is making some headway into trying to untangle this web of communication, by looking at twitter data from the UK over a period of a few years. This consisted of several million tweets, analysed by examining the words to identify the sentiment expressed by the tweet. While one individual tweet is not particularly meaningful, taken together clear patters begin to emerge, and correlate strongly with real-world events happening. For example, the 'joy signal' - a measure of how happy people on average are according to their tweets - reliably peaks each Christmas, as does the 'fear signal' at Hallowe'en.

So far so predictable. But a really interesting thing happened after the pre-budget announcement in October 2010, when the first wave of spending cuts was announced by the new Conservative government: the national level of fear shot up, as people immediately began to worry about losing services, benefits or livelihoods. What's remarkable about this is that the fear level, after an initial spike, did not go back down - it remained at an elevated level until the end of the sample period, deviating noisily from what looked like a new baseline. A similar thing happened in mid 2011, when riots broke out in many UK cities, and for a while the anger signal dominated. This gradually diminished as the rioting died out (back to the baseline fear over continued cuts). The researchers, of course, were quick to downplay suggestion that this could be used to predict future riots.

Whether the sampled data was actually long enough to show this is a long term change, or if such drastic changes in average mood are common, is not yet clear. Either way, such analysis promises to be a useful tool in understanding not only how people are using their own media generation abilities, but how they react to wider events. I am reminded of a plan by the government a year or so ago to measure the nation's happiness, and to use that as an indicator of prosperity (arguably makes as much sense as other arbitrary measures like GDP) - this would be a useful alternative, requiring no interventionist surveying, and giving a more immediate picture of what's going on. That this kind of information can be gleaned from tweets alone, using a fairly simple model of content understanding, is quite encouraging, given the wealth of data which more sophisticated tools could exploit.

I think that, despite the concern over what people are reading and how they see the world, or the effect that new media may be having on our perceptions, there is grounds for optimism. On the one hand, it seems inevitable that people will gravitate towards the media they prefer, in turn reinforcing their entrenched view of reality - and with a better model of this, ways to target them more precisely will emerge. But with this knowledge will also come the possibility of disentangling the complex web of media interaction, figuring out why people seek the information they do, and how to use this for the improved dissemination of information. User generated content will only become more important in how people interact with the world - but rather than being a cause for concern, this can be a valuable source of information, to directly measure how they perceive reality, and how people respond to events. The media as it stands is a vast and complex system which, much like the world economy, is beyond the comprehension of most humans: but with more sophisticated tools in data mining and artificial intelligence, it may be possible to get a better understanding of how this all works, and how we should deal with it.

Justin Lewis, speaking on climate change and the pervasive advertising industry:

Interview with Nello Cristianini on BBC news about the twitter sentiment analysis project: