We interviewed the author of "Information Foraging Theory: Adaptive Interaction with Information" about his new book and the theory behind it.
Website Optimization (WSO): What is the purpose of your book? Who is the intended audience?
Peter Pirolli: "The book is aimed at an interdisciplinary audience - people who might have backgrounds or interests in the application of cognitive science, computer science, information science, and economics to human-information interaction. The emphasis in the book is on theory and research, but the goal of that scientific work is to get to a deeper understanding of what people really do - and why - when they're trying to find information to solve their very real everyday problems. So, the book also tries to pull out key insights about human information foraging, such as the crucial role of information scent, as well as presenting design principles, engineering models, and novel user interfaces that were motivated by the theory."
WSO: Can you summarize what you mean by information foraging?
Pirolli: "Any behavior that involves seeking and consuming information for some purpose - so that's pretty broad. The use of term 'foraging,' however, is purposely chosen to draw upon a particular metaphorical way of looking at the world (a particular 'paradigm,' to use that overused word). I think that web designers and user experience researchers resonate to the idea that people have to navigate through information spaces following clues to the prey they are stalking; changing their strategies and technologies according to circumstances. It seems easy to envision mental models of animals foraging and then form analogies to what we see when we watch people surf the Web. The bonus of the food foraging analogy is that there is a whole scientific food-foraging discipline (optimal foraging theory) devoted to developing quantitative models that look a lot like economics and engineering. It turns out that these quantitative models transfer quite well to understanding human information interaction."
WSO: What is satisficing?
Pirolli: "'Satisficing' is concept introduced by Nobel Laureate Herbert A. Simon that pretty much revolutionized economics. The classical 'Rational Man' concept assumes that when people are making decisions they use all the information about the alternative choices and choose the best. Simon's big idea was that people actually operate under 'bounded rationality' - there are limits of time and resources that have an enormous impact on decision making. Before the Web, buying a car was a real time consuming process, and most people certainly didn't have the resources to go to every location to collect price and quality information. That same thing applies to the Web, whether searching for a good buy or the best possible answer to a question. We just don't have infinite time and resources. So we're blocked from absolutely 'maximizing' our decisions and instead 'satisfice': We use rules of thumb in setting aspirations about what to achieve in what time with what resources."
WSO Can you explain this what spreading activation means, and how it applies to information foraging?
Pirolli: "Spreading activation is a scientific technique for modeling how human memory works. The general idea is that the mind associates ideas to one another according to psychological laws. A whiff of perfume reminds you of your girlfriend, then the occasion she wore it, the taste of the wine you had, and so on. Some cue (perfume) activates a dormant idea in your head, then that idea activates another, and so on. The notion that ideas associate with one another in particular ways goes back to Aristotle, but it is only in that last 35 years or so that we've been able to actually develop a formal, mathematical theory of how it actually works. Spreading activation is the computational technique that allows you to say how cues in the world evoke ideas in the mind, and how each idea activates associated ideas. In the context of the user interfaces, we're interested in how cues on something like a web page evoke ideas in the mind, and then there's a little bit more theory involved that specifies how the user makes judgments based on the relation of those evoked ideas to their information goals. The beauty is that we can build automated engineering models that have huge models of human associative memory embedded inside of them that make predictions about human judgments about cues on a user interface."
WSO: You write that starting with a high information scent is strongly associated with longer website runs. How do you define high information scent? What is the average? How does one calculate scent?
Pirolli: "Information scent is calculated using the spreading activation models. First, it's important to note that information scent scores are computed based on the amount of spreading activation between cues on a UI and the user's information goal. So it's a relational measure that varies from one user goal to another. Second, when we say 'high scent' about the initial home pages encountered, we mean high relative to other home pages of other web sites. So again it's a relative measurement. Unfortunately there isn't a standardized scale for information scent measures yet. There used to be a bar of platinum stored in very regulated conditions in Sevres that corresponded to 1 meter and the whole metric system was founded on that standard. Maybe what we need is a standard user goal and web link stored in controlled conditions in the basement of the National Institutes of Standards and Technology and build up a whole information scent measurement system on that."
WSO: You seem to imply in your book that information scent is more important than aesthetic design, is it?
Pirolli: "I think it's impossible to say one is more important than the other. I think we all love things where beautiful form and efficient function come together as one. Having said that, there are so many UIs and web sites where one immediately gushes 'oh that is so cool' and after using it for a while you realize that it just doesn't do the job. We've developed a few things at PARC that were like that. So what I was saying was that in terms of functionality - getting people to the information they desire - one has to focus on information scent as opposed to aesthetics. But it is also interesting how good function can often become aesthetic or cool. For instance Marissa Meyer recently attributed Google's user experience to good information scent, and Google is pretty cool these days. See:"
WSO: Backtracking: You write that low information scent can increase the cost of foraging from linear to exponential after the "false alarm factor" exceeds 0.1 (10% of wrong paths taken) for branchiness factor of 10. It appears that the cost of navigation can be quite high in deep sites with low scent. Galletta et al. (2006) found that deep unfamiliar sites were nearly 7 times slower to navigate with users than broad familiar sites (see below):
Pirolli: "That example is a hypothetical case used as an illustration of a general relation between the branchiness, depth, and information scent of a site. In that case, I was assuming that the hypothetical site had a branchiness of 10 - e.g., 10 links emanating from every page.
Let me try to explain without the math. At a perfect web site, the user would never choose the wrong path (the "false alarm factor" would be zero). Even on a deep web site, the user would make only the clicks necessary to get to a deep location, regardless of how many other links choices were available on any page along the way (i.e., regardless of the 'branchiness'). But now, suppose that on average a user was likely to choose 1 wrong link. Now people are likely to have to back up on their way to their targets, and deeper sites are going to compound the number of backups. The deeper the site, the greater the compounding. Roughly, if you have a probability P of choosing a wrong link and there are on average B 'wrong' links on a page, then as soon as you go over P*B = 1, you're likely to have this compounding, exponential, web navigation cost. So increasing f (poor information scent) or B (more poor links per page) can each have an impact. I think what Galletta et al. found is this effect in the real world as opposed to the hypothetical case I was discussing."
WSO: When I research a topic, I tend to go to Google Scholar or a university search system to increase the richness of the patch I'm searching. Also, limiting searches to PDF documents can help. Do people naturally gravitate towards refining their searches?
Pirolli: "We see this everywhere, in studies of people working on research papers, intelligence analysts, journalists, and so on. There's a trade-off between exploration, enrichment, and exploitation and we typically see people cycling through each of these processes. Exploration is a kind of general assessment of what's out there. In the intelligence community, for instance, there's a lot of incentive to make sure that one knows about every possible bit of relevant information. Enrichment is a process of somehow eliminating the stuff that is likely to be of low value - along the lines that you've mentioned. Exploitation is then the actual work required to extract out and make sense out of the information you've collected (which takes up time and resources too). To give you a real-world example, we studied MBA students working on business strategic analysis papers, and their Exploration would involve using the Web, library catalog, and Lexis-Nexis to get an initial listing of what was available. Then their Enrichment would consist of a kind of quick triage based on titles in their listings. They would also trade-off things like length - for instance books would be too long to read given their deadlines."
WSO: What can webmasters do to increase their site's information scent? Got any guidelines or rules of thumb?
Pirolli: "I have a bunch that I've collected from various UI gurus who have been working with the information scent concept. Those are in the book."
WSO: Cooperative information foraging: Web 2.0 seems to be based on this concept, where multiple people attack a problem and find a more thorough answer or solution. What examples can you provide of cooperative information foraging? Is the optimal group size 7? What do you think of wikipedia.com versus the Encyclopedia Brittanica (two different models)?
Pirolli: "Two interesting books about this (in addition to mine of course) are James Surowiecki's 'Wisdom of the Crowds' and Cass Sunstein's 'Infotopia.' Wikis, social bookmarking sites, predictive markets, and the scientific literature are all examples where people cooperate to aggregate information, make discoveries, solve problems, etc. In many cases, cooperation can produce better answers than any of the individuals - even if the individuals are highly expert. The problem is, groups can also work very badly - repressing important information held by some members or amplifying cognitive biases.
One way to understand Wikipedia is that it has many characteristics of open source software. It is modular, so components can be added and 'debugged' more or less independently of the rest. It is very easy for 'experts' to see where they might contribute or correct something. There is a thin layer of administration and editorialship to make decisions and clean things up. It relies on the fact that when you have millions and millions of minds out there, someone with even small amounts of knowledge about a topic will have sufficient motivation to see their words in print, and lots of other people will be motivated to correct them. So, the medium includes mechanisms for aggregating knowledge, combined with mechanisms for editorial deliberation and correction. Recent evaluations show that Wikipedia is remarkably good, though not perfect, when compared to Brittanica, and of course it is very uneven in coverage. However, I can't remember the last time I used Brittanica and I use Wikipedia every day."
WSO: Cooperative foraging - can you explain how this figure works with web surfers? (see Figure 8)
Pirolli: "The numbers are all hypothetical, so no claim is made about any particular domain, but there are a few general points, Even if cooperation is a good thing for some task, cooperation also tends to have various kinds of costs (e.g., communication costs). Also, in many cases, the returns to the individual diminish as more and more people join in. There will tend to be an optimal size to the group (the peak in Fig 8), and if you are a member of the group, then that's the size that you would prefer. But, if you were outside a group that was reaping the benefits of being at that optimum, you would still want to join the group even if it diminished benefits slightly from the optimum (i.e., one more person would join the group, the rate of return to the members would take a bit of a hit, but as a "new joiner" it would still be way better than not being a member of the group). So you tend to get a conflict between "members" who would prefer to keep the size of the group at an optimum, and 'joiners' who would benefit from being in the group even if it isn't at the optimum. At some point the tension between those two tendencies balances out, and this equilibrium point will tend to be at a group size that is larger than optimum.
So, fishing fleets tend to adopt secret codes so that fishing information can be exchanged (cooperative foraging), but the group size can be controlled. I think we tend to see this sort of thing in forums (how many have you seen dissolve into uselessness with increases in group size) and perhaps collaborative tagging/bookmarking systems."
Peter Pirolli, the father of Information Foraging Theory, has written a landmark book about a new field in human-computer interaction. He challenges us with formulas, diagrams and bold insights to learn and adapt, or be left behind. In our world of information overload, knowing how people find their way and how we can help is critical. Designers who foster adaptive strategies to allocate attention efficiently will be rewarded with website success.