[I'm republishing some of my favorite articles of 2009.]
Is it possible to describe aspects of human behavior through algorithms? Hewlett-Packard's Bernardo Huberman believes you can and he has the studies to prove it.
They look at fairly ordinary activities such as the number of YouTube downloads, the number of "Diggs" a web page receives, or the number of times people upload and share content. From such mundane activities his team can derive algorithms that seem to uncover aspects of human nature and provide a glimpse into things about ourselves that could be universal, that could very well be possibly hard-wired into our very being.
The key to this research is that the studies are extremely large, the sample sizes are in the many tens of millions. The larger the samples, the clearer are the patterns of human behavior that emerge.
"We did a big study of 70 million YouTube video downloads and also millions of Digg stories. From that we can now tell which content is about to go viral," says Mr Huberman.
The study is part of the Social Computing Lab's work on what holds an Internet user's attention.
"Attention is a limited resource. We cannot produce more attention. It will be forever limited."
The study showed that attention is a function of novelty. And that attention will decay in a predictable pattern as novelty decreases.
"We see the same graph, time and again. All content has the same graph of attention and decay."
The shape of the graph is not a "Huberman distribution" it is one that has been familiar to statisticians since the mid-19th century, it is a lognormal distribution, also known as a Galton distribution.
Since the popularity of all Internet content always exhibits the same curve it is possible to make judgements about which content is rising in popularity and therefore promote it on a web page while removing content that is falling in attention.
Catching your eye...
HP has developed a tool it calls i-catcher that can be used to maximize attention for any web content. The tool is currently being tested at a large Danish publication where it helps determine the position of content on the page.
Although such algorithms work best when used against large amounts of content, Mr Huberman says that i-catcher has also proven effective within HP's internal blogs, which only have several thousand visitors.
Mr Huberman's team has also studied uploading of videos, numbers of comments, ratings, etc. Not surprisingly, the more of this type of this involvement and sharing activity the higher the quality of the content. But surprisingly, the more people uploaded the less successful was their content in terms of popularity.
What is people's motivation when they upload content and share? Mr Huberman says it is based on a strong desire to gain attention.
"We are attention machines, we constantly crave attention."
But not everyone is a Paris Hilton or a Julia Allison. Clearly, many people are able to satisfy their relatively modest hunger for attention in moderate ways.
Still, this craving for attention is anther insight that Mr Huberman's team has uncovered when it comes to predicting the mass behavior of Internet users.
Another interesting quality of attention is that it is asymmetric. I ask what this means.
"People will give attention to content regardless of the quality of the content. People are attracted to popularity. They are interested in what is popular and that is often independent of the quality of the content."
I ask how these findings are being used within HP. For example, the marketing department could use these algorithms in their efforts to gain attention for their marketing messages, and to sell more HP gear and services.
Mr Huberman smiles," Yes, marketing is very interested. It seems that for the first time, the work at HP Labs is relevant to the marketing department. Labs has never worked with marketing before."
- - -