The study of the field of the description of tasks and adaptive interfaces has allowed us to identify four objective variables which are independent of the subject and make it possible to characterise an information-seeking task. These variables are "the repetitiveness of the task", "the level of explicitness of the targets", "the location of the targets" and "the quantity of targets".
The study of adaptive interfaces has shown us that intelligent agents are useful for repetitive tasks. These are tasks that are performed on a regular basis which require the use of the same operations in order either to retrieve the same information or a similar piece of information. They are distinguished from non-repetitive tasks which are one-off tasks
Ever since the first empirical studies of the use of hypertexts (for example [ROUET, 1990]), researchers have distinguished between explicit and implicit targets. An "explicit" target corresponds to an extract from a document, for example a text paragraph, which simply has to be understood by the subject. The subject does not need to search for any other information or produce any inferences to attain the goal. In contrast, an "implicit" target requires the subject to call on additional information or knowledge in order to achieve the goal since the target is not sufficient in itself.
On the basis of an analysis of a variety of empirical data, Tricot [TRICOT, 1993] has defined two other objective variables which make it possible to specify a search task. These are the quantity of targets and the location of the targets.
A target's location is the place at which it can be found on the information networks. Here we distinguish between precisely located and distributed targets. If a target is local then it is fully present on a single page (in certain cases, it may also be redundant, i.e. it may be present "in the same way" on several different pages). If a target is distributed then it is present on several pages and the subject must view all these pages in order to obtain all the information. If the target is "local" then the search terminates as soon as the site containing the target is found. In contrast, if the target is "distributed" then the search has to be continued if all the information is to be retrieved. The search for "distributed" information may take some time and imply a greater cost.
The "quantity of targets" variable defines the number of targets that exist for a search task. For this variable, the information is fully present on each of the pages on which it is present but there may be one or more pages or sites containing this information. If the information is "unique", it is only present on a single page and the difficulty for the subject is finding it. Otherwise, if the target is "multiple" the subject has a number of equivalent ways of finding the target.
We want to assess the effects of these variables on users' activities. We imagine that it will be more or less difficult for subjects to perform information-seeking tasks as a function of the values of these variables. The adaptive and customisable interfaces as well as the new media (mobile, PDA) correspond to specific utilisations. We imagine that the study of the effect of our objective variables on the search tasks conducted on the Web will enable us to arrive at a partial description of these utilisations.
To do this, we have defined an experimental protocol by means of which we evaluate subjects' performances when performing information-seeking tasks defined on the basis of our variables. In the multimedia field, assessing subjects' performance is not an easy task since the behaviour of one and the same subject may be judged to be good by one author but not by another. Nevertheless, there are rational criteria which are acknowledged to permit a good assessment of subjects' performances. These are the recall and precision indexes . Recall evaluates the number of targets attained by the subject out of the total number of existing targets. When the recall index has the value one, the subject has found all the targets. Precision is the ratio of the number of targets found by the subject to the number of pages opened. If the subject has opened only relevant pages during the search then the precision value is one. We evaluated subjects' performances on the basis of these two indices.