In a short article, the Register reports on an update that Bernardo Huberman, a senior fellow and director of the Social Computing Lab at HP Labs, gave to a room of journalists yesterday on what the group has been doing.
One of the topics of research was on how a web site can more effectively present its content.
“After studying about 1,000 digg.com news stories, HP said the team was able to make a mathematical model to predict how long it takes for the popularity and novelty of an article to die off and disappear from the front page. […]
Huberman said their model suggests that arranging a web site so that new and novel items are most prominently displayed is generally more effective at attracting clicks than prioritizing based on its popularity.
HP said it tried the process on its own web site to select which items are recommended to customers. Preliminary results, according to HP researchers, showed a 30 per cent increase in the attach rate of sales.”
Interestingly, the researchers also found that “user recommendations generally yield surprisingly unimpressive results”.
“As people become increasingly resistant to traditional forms of marketing, viral advertisements and recommendations are being used more often as an alternative strategy.
But the lab found that while the likelihood of someone buying a product does increase with the number of recommendations at first — it soon plateaus to a constant, but relatively low probability of purchase.”