07 Apr 1997 ............... Length about 1000 words.
This is a WWW version of a document. You may copy it. How to refer to it.

TOPIC 1: "Push" of information by supplier vs. "pull" by a user

We are familiar with a contrast between browsing and goal-directed retrieval. Much the same contrast applies at (at least) two levels: the user's prior goal or "information need", and the method the user employs for satisfying the goal which is strongly constrained by the mechanism supplied. Goal directed retrieval is when the user knows exactly what they want e.g. the fax number for sending a fax to a specific person. A browse goal is when a user looks through information (e.g. an individual's web page) just "to see what is there", or "for something interesting". A browse method is when users scan rather than issuing a precise request: it is most directly supported by hypertext or flipping through pages. Directed retrieval is when the user issues a specific query e.g. in SQL or typing in keywords to an IR system. Of course users may, or may have to, use either method for either goal e.g. if I want someone's fax number I may have to browse their own web page, then that of their department or university; or conversely, I may issue a keyword such as "HCI" to a system just to see if any interesting looking papers come back, even though I put nothing in the query to define what I would count as "interesting". More advanced interactive systems also may mix methods e.g. the Rabbit system used aspects of the browse method for directed retrieval from databases, and Ian Campbell's software uses aspects of the browse method for directed retrieval from IR document collections.

Another distinction concerns how much the author (information provider) takes responsiblity for the selection. In IR and databases, they take no responsibility, but a lecturer takes all the responsiblity and demands that students turn up and listen to whatever they say. This may seem to be a consequence of the medium: since the lecturer can only speak about one thing in a given time slot, someone has to decide what that is: it cannot be left to the individual student. However this is not really the case, as is shown by hypertext. The main argument for hypertext was often that it allowed the non-linear (network) structure of most material to be directly represented, instead of forcing an artificial linearity as in lectures and books. When early hypertext authors responded by putting up network structures, it became clear that this was often not what users needed. This made clear, particularly in teaching applications, that an author has two different jobs: one is to create the material with its inherent network structure, and the other is to create one or a few recommended "narrative" routes through it. Although when you are using a textbook for reference, you do not need these guided routes but will use the index and contents to do goal-directed retrieval, a new student does not have the knowledge to do this but requires the author to provide a good route to follow. This is not browsing: the serious student will try to absorb it all, not skip through to pick out a few interesting pieces. Instead, this is a case of goal-directed retrieval where the author (information provider) making most of the choices for the user: reducing the load on the user, and providing an important service.

This is seldom discussed in IR. It is a crucial service, as the failure of hypertext without it shows. It amounts to vastly reducing the user's choices, and hence their workload; of course it is difficult to do this completely satisfactorily. Note that it does not remove all choice: users (students) still choose whether to come to the lecture, which book to use, whether to stop paying attention; but they no longer have to make continual route choices. In education it is particularly important, and is why extreme learner-centered models are misguided: teaching and learning is more like a dialogue, with both parties needing to carry some of the burden of choice. However it applies much more widely. As is becoming clear with the advent of electronic web-based academic journals, one of the functions of a journal is to provide quality control through refereeing to cut down the number of papers published. This is not just to save publishing costs, but to save readers reading time by only offering the better papers. Readers are paying partly to have the editor make reading choices for them. Similarly with broadcast television (and magazines and newspapers): the broadcasters are not just reducing the volume of material to fit into the broadcast channels available, but also making choices for the viewers. This is a valuable service, although of course it depends on how well they make that choice for you. To take a paradigm example: a person may watch one 30 minute TV news programme every day. They have no control over content, and because the length is fixed, the quality (importance and interest of the news) is actually very variable from day to day. Nevertheless, it saves the user from making any effort: they just do the same action every day (switch to that channel at the same time). This is the highest usability (lowest cost) of any information interface possible, but the utility (i.e. benefit) is of course questionable and non-optimal. It is perhaps the extreme form of a browse method for a browse goal.

The advent of hundreds of broadcast TV channels (digital channels, terrestial broadcast, satellite, cable) is likely to re-create the problem: I suspect that that amount of choice will not be usable unless new search mechanisms are provided, or alternatively most individuals will just stick to one provider and trust them to make the selection. Apparently Microsoft is going to apply this to the web, and provide pages that change constantly so that navigation for (trusting) users will not be a problem: new content is provided constantly to entertain. In this situation, providers end up having most of the control over what the user "needs" to know. Will their choices serve users interests or only the providers' interests? We give them that power however because they are doing us a very big service in removing the work we would otherwise have to do.

It would seem, then, that the subject of information provision and use is no longer tied to the medium: we should be looking for concepts that apply equally to databases, the web, and broadcast TV channels. In evaluation and other design-related work, we need to classify the different types of user value e.g.:

Summary

Implications for evaluation

This stresses the importance of studying (evaluating) undefined or tacit user information needs such as to be "interested" or "entertained".

It suggests that we might want to use measures, not just of total user costs in terms of time, errors, mental effort, but also some measure of the division of labour between user and provider. This would scale properly as the information space got larger.