Andreas Dieberger
Emory University / ITD
540 Asbury Circle
Atlanta, GA 30322
andreas.dieberger@acm.org
Kristina Höök
Swedish Institute for Computer Science
Box 1263
S-16429 Kista, Sweden
kia@sics.se
A shortcoming of the World-Wide Web is that users get the perception they might be alone in information space. There is a strong trend to more populated information spaces that increase one's awareness of other people's activities. This awareness is beneficial to judge the utility of certain information, its recency and also its rate of change.
We describe modifications to a collaborative Web server (SWIKI) that enhance the user's feeling of being part of an active community on the Web server. The modification try in part to extend the server into a social navigation system, in part try to experiment with how people react to improved awareness of other people within the information space. We found that our modification give a good feeling for the change of information in the site and help regular visitors to navigate to newly modified information and to information of current interest. We further describe how social navigation principles can be applied in the design of an online store to buy food. Especially food stores tend to be very social places and the design of virtual food stores needs to incorporate social aspects much more than online stores typically do.
Although many people may access a Web site at the same time, the Web mostly maintains the illusion of dedicated resource. The only indication of a larger number of users simultaneously accessing a site might be an unusually slow response time. Although humans are very social animals, most of our social skills are ignored and unused on the Web. Virtual communities and chat systems are only small pockets of social activity in a sea of anti-social information spaces.
Navigation is a social and frequently a collaborative process [7]. A relatively recent line of research, social navigation, tries to find ways to better use our social abilities in virtual information spaces. Social navigation can have the form of direct social navigation, where people give recommendations or guide each other, or it can be indirect. In this second form of social navigation the system creates a feeling of awareness for other people's activities in an information space. Information on where people frequently went and what people found interesting, aggregated over a community of users, is a valuable tool to make navigation decisions Ð especially when this community of users has shared interests, see [1] or the forthcoming volume [8].
Empathy is basic to human thinking. Usually we assume that other human beings, animals, and sometimes even inanimate objects, are intentional creatures with emotions and interesting relationships to us and other beings. A large part of what we talk about during the course of day concerns discussing why people have performed certain actions, or what their intentions could be. We try to figure out who are still our friends, who is friends with whom, and we re-establish our position in the ranking scale. Some researchers, e.g. [3] actually claim that this social "grooming" and its evolutionary advantages caused the development of the large human brain and complex language.
The success of chat rooms and Web bulletin boards shows that people feel a need to interact and to share thoughts and information. Social navigation processes are very common in everyday life. The number of cars parked in front of a restaurant is an indication for its popularity as is the length of a waiting line before a theatre. We frequently base the choice of a movie or restaurant on friends' recommendations or on articles written by well-known critics. Few people would choose a dentist based solely on the listings in a directory of doctors. They rather get a personal recommendation from somebody they know. The same is true for buying cars, selecting books and sometimes even meeting people. All of these processes are variants of what we call social navigation.
What is interesting for other people is probably of some use to us, especially if these other people are in a community that shares certain interests with us. This reasoning forms the basis for recommender systems, which are a branch of social navigation systems [11, 13]. Furthermore we know that recency of access to information is an excellent predictor for future access [9, 10]. It is reasonable to assume that a system that actually shows what information is accessed can improve a group's focus on documents or general interest.
These and similar observations motivate research in social navigation. In the remainder of this paper we describe two systems that are designed around ideas of social navigation. One of them is a collaborative Web server, and the other is the design for an online store for food.
As first example we describe modifications to a collaborative Web server, called Swiki, standing for Squeak-Wiki. It has been implemented in Squeak (a free dialect of Smalltalk) by Mark Guzdial of the Georgia Insitute of Technology. Swiki is based on earlier work by Jim Cunningham (Wiki Wiki server) and influenced by Tim Jones's WebTalk system. (For more information on Swiki see http://www.cc.gatech.edu/fac/mark.guzdial/squeak/pws/.)
What prompted our use of the Swiki is its openness to modifications (open source) and the fact that every user can easily modify and extend every page on a Swiki. To do this, a user simply clicks a link called "Edit this page" and modifies the page in a Web form. Pages consist of plain text, but people can use HTML if they like to. Links, new pages, embedded images etc. are created by simply surrounding a URL or a page name with asterisks. The ease with which people can contribute on a Swiki server makes it an ideal tool to support collaborative writing, open discussions etc.
Swikis support authorization to protect pages, but we decided to leave the server completely open (like most Swiki servers). One might fear that a completely unprotected server would get easily be abused by hackers. In practice we don't know of a single reported case of hacking or posting inappropriate content on a Swiki server. A reason for this could be that there is no challenge in breaking a Swiki.
We describe a Swiki modified by the first author. Among the modifications is a different user interface of the Web pages. We also eliminated several standard Swiki features to further simplify the system (for example there is no search). Instead we added functionality to enhance the awareness of the community's activities.
All Swiki servers maintain a list of recently modified pages, however there are many more consumers than producers and this list doesn't give a good indication of the activity of the community. (In addition to having more consumers, we (and other researchers, for example Mark Guzdial, pers. comm.) found that people are quite hesitant modifying other people's Web pages.) We provided an additional function that points out modified pages in every link. Hand-coded "new" markers appear on many Web sites, but they seldom give a good indication for the newness of the information. On the Swiki we instead have 3 levels of newness. Pages that have been modified within the past 24 hours are pointed out with a different marker than pages that have been modified within the past 3 days or pages changed within the past week. In addition, we point out links that indirectly lead to new or modified pages (2nd level what's new marker) and thus point out paths to new information.
The essential difference between our "new" pointers and a global list of changes is that new markers on every link show modified pages in a local context. In social navigation information stands not isolated but always has to be seen in context. As the system maintains the markers automatically it is guaranteed that the markers are all up to date and show correct information, wherever a link is places. Therefore these markers are visible also in the list of "recent changes".
Originally we used 4 levels of newness and pointed out pages that were modified up to 2 weeks ago. We soon discovered that a fourth level of newness didn't add to the system, but was even distracting. We also found that recency information beyond one week was essentially useless. However, each virtual community is different and its level of activity changes over time. Our simplistic scheme of newness markers does not take this into account yet. Future systems need to adapt to the level of activity and the rate of change and adapt the markers appropriately. Ideally the system should point out a page as new only if a user has not seen that latest version yet.
The list of recent changes allows users to quickly find modified pages, but it doesn't give an overview of the general traffic on the site. This information is visible in two ways: we implemented a list of "recently accessed pages" and we use symbols of footprints to point out links leading to pages that have been used within the last 24 hours. It is important to stress that footprints are not a property of the link but of the page. A footprint thus signals that the page got accessed, but not necessarily through the link at question. This is a significantly different approach than for the read wear on MOO room exits as it was used in the Juggler system [1]. A Swiki footprint shows aggregated usage information for pages over all possible paths to that information.
We again chose 3 different levels of footprints. Ideally these levels again should be adjusted to the community. For pages that had no access for over a week we show a little dinosaur. At first the dinosaurs were implemented mostly as a joke, but we soon realized that they are an excellent tool to immediately see whether a community has more or less died out or whether certain subjects are of no interest any more.
We found that the footprints and dinosaurs give a pretty accurate feeling for the activity of a virtual community. We decided to point out accesses within the past 24 hours based on [10]. Pitkow and Pirolli report that Web pages that have been accessed within the past 24 hours are likely to be accessed again. One of the goals of our work is to study if and how much the likelihood of page access changes when this recency of information is actually made visible to the user. In the long run however, the time constant for footprints should be adjusted to the level of activity in a community as mentioned above.
A further feature on our Swiki server tries to create short-term awareness of activity on the server. This feature, called "What's going on", points out accesses and modifications within the past 5 minutes and which pages are currently being modified. We do not have much experience with this new feature yet, but this snapshot of activity seems to be most useful in times where several users use the server at the same time. A scenario for the use of this feature would be a Swiki meant for use in a classroom, were all students might check the Swiki right at the beginning of a class or at designated lab hours. Swiki servers have also a chat room feature, that ideally would be coupled with the "What's going on" page. On our simplified Swiki we don't use the chat page.
Like many social navigation tools, our modifications of the Swiki seem to be most useful within a small closely knit group of users, for example a small work group, where people are aware of each other's interests etc. Our next example instead looks at a larger community, where people are likely to be unknown to each other.
While the Swiki example is focused around a collaborative environment where social trails should be a natural ingredient, we would like to claim that social trails are useful also in other, less obvious, environments. One such domain is on-line food shopping. Existing food stores on-line are typically 'dead' spaces where users use a form to specify how many milk packages, etc. they want sent to their doorstep. There is no indication of other users and the feeling of a social place that typically is strong in physical food stores is completely absent. In a study on shopping in a VR environment [12], it was found that users in virtual stores like the social aspects of a physical store, that they wants to socialize with other people also in virtual stores and crave a multi-user experience. Shoppers in an on-line store in Stockholm frequently start by traveling to the corresponding physical store in order to get a better picture of the items. They also want to get an impression of the store's personnel Ð probably since this makes the on-line store come alive with imaginary people Ð it is no longer a dead space.
Using a 'traditional' view of human-computer interaction view we could approach the design of an online food store as an interaction problem. We might ask how to make it easy for users in front of their home computers to find the items they need, 'pick them up', put them in their shopping basket, pay for the items, and make sure that they get delivered at the right time? Underlying our worries would be whether the interaction is designed optimally to be as efficient as possible from a user's point of view, down to the level of the number of actions to achieve a given task. We might also consider the aesthetic aspects: are the items displayed in an inviting way? Is the experience of walking through this information space a pleasurable experience? Is the right metaphor chosen, etc.? While all of these design considerations are highly relevant and should not be taken lightly, they are all based on a one-computer-one-user view of interaction. Food shopping is a very social process. Using social navigation features we try to socially enhance the virtual store.
First of all, we assume that other people are "around" in the store. Instead of imagining a "dead" information space, we now see before us a lively space where (in some way) the user can see other shoppers moving about, can consult or instruct specialist agents and 'talk to' the personnel of the grocery store
Perhaps more interesting is the question how to accumulate trails of people in the store to visualize community preferences or recommendations. In the context of the food store, recipes can function as accumulated pieces of knowledge. Through our choice of recipes Ð which in turn influences our shopping list -- we convey a lot about our personality, which culture we belong to, our habits, etc. Making recommendations on which food to buy based on recommending recipes is an interesting functionality in itself. Imagine adding accumulation of user behavior so that we understand which groups are most likely to choose which recipes.
In the online food store currently being designed at SICS a user logs onto the system, and will get a recommendation for a recipe, based on popular downloaded recipes in the community for category of people the user is in (for example vegetarian). The user can add the recipe to his/her shopping basket, which in turns adds the ingredients from the recipes to the list of items that will be delivered to their doorstep. The user can ask for the next-best recipes that fits with his/her category of users Ð much in the same line as Amazon.com recommendations ("other people who bought this book also bought these books").
The recommended recipe will be chosen on the basis of three different characteristics that the user can manipulate: the group of users that the current user belongs to, the category of food (Italian, Thai, etc.), and any particular ingredient that should be included (shrimps, beef, etc).
The problem with such recommendations, and recommendations in general, is that they provide the user with very little insight into who else is in her group why she was placed there. Our solution to this is to put an "editor" back into the loop. The editor will look at the clusters of users (based on which recipes they have chosen) and "name" those with fuzzy names that conveys somewhat of their content: "vegetarians", "light food eaters", "spice lovers", etc. The user can then in turn choose to have a look at the recommendations done for another group of users that s/he does not belong to. This way, a user can try out being a vegetarian for a week. It also provides some insight into the inner workings of the recommender system [6].
The shopping system will also provide comments (and discussion of recipes), an optional image of the person who contributed the recipe, an image of the dish etc. to provide richer social trails on top of the recipe. Our solution will provide the users with more insight into both the social trails of their own actions as well as other users' actions that have lead to the recommendations they finally get.
The work reported in this paper touches only a small facet of the relatively young research area called "Social Navigation". Despite the name, social navigation is not only concerned with navigation proper, but with all forms of decision making that are based on the behavior and activities of other people or groups of people. One is tempted to put social navigation into one basket with recommender systems, but this would be a gross oversimplification. Social navigation is a wider concept than recommendation based on user profiles or personal preferences. However, recommendation based systems, like Amazon.com's book recommendations and earlier work on recommender systems, for example [13] are important aspects of social navigation. Other work that is directly related so social navigation is the study of how people perceive spaces and places [2, 4], see also Dourish's and Dieberger's chapters in [8] or work on visualizing histories, for example [14] and the original work on history-enriched information spaces in [5].
An interesting aspect of social navigation is the aggregation of information as it occurs in the Swiki footprints. It helps avoiding some of the issues recommender systems runs into, in particular issues of trust and privacy. For more information see Dourish's chapter in [8]. Another aspect of social navigation concerns the design of the environments for virtual communities and the difference between virtual space and place. The design of a virtual community can shape opportunities for action and understanding for permissible or frowned upon activities in a shared space. This has obvious impacts on systems like the Swiki and an online shopping mall because it suggests permissible and unaccepted behavior, and again gives context to the activities of the user population (see Dieberger in [8].
We see a wide range of application of social navigation features as described in this paper. They give not only useful information for making decision, but create a sense of community, and often belonging. They therefore can inspire people to return to a system (which has obvious implications for online commerce). Besides social navigation's utility in shopping systems and small, closeLY knit cooperative communities, showing usage within a community can also be useful in education systems, or any system where certain members of a community might act as trailblazers through information. Instructional systems might show such "read wear" only of people who are considered experts. The system thus would teach how to best navigation the information space, without restricting deviating from an established "path".
We mentioned the importance for context for information. Information on who is accessing information and who isn't can be as important an aspect in a navigational decision as the validity of the information itself. For example Harper reports an example in his chapter in [8] where an information source, despite providing superior data was not considered useful, because the 'people who count' didn't use that particular resource.
For the close future we see a host of more socially enabled systems appearing on the Web. We have reached a point where people want to interact in social settings on the Web, where online resources become "places" instead of being sterile information repositories. Today's systems have reached a level of sophistication where they can be communication tools and interfaces to socially interact with people near and far. Social navigation processes are an important ingredient for making systems true social places.