5 Discussion


5 Discussion

The results that we obtained permit us to conclude that each of our variables had an effect on the information-seeking tasks . The greatest effect was exerted by the "repetitiveness of the task" and "explicitness of the targets" variables while the "target location" and "target quantity" had a lesser effect.

We observed that when subjects successfully perform the first task in a series of repetitive tasks, they apply the same search strategy in the succeeding tasks which they also perform successfully. Recall and precision for each of the "repetitive" tasks are therefore high. In contrast, we observed that a subject who failed in the first task, performed the second and subsequent tasks on the basis of the results obtained in the preceding tasks and either changed their strategy or re-used the same strategy in an attempt to improve it. This results in an increase in recall and precision scores. We may well imagine that the fact of repeating the task allows subjects to improve their searches through a learning process. We may therefore hypothesise that repetitive tasks are tasks that are frequently performed successfully. This would seem to call into the question the utility of adaptive interfaces for information-seeking tasks. Nevertheless, repeating the same actions or sequences of actions is time-consuming to users. We might therefore imagine that adaptive interfaces which automatically perform this job would prove to be very useful in that they could enhance ease of use by eliminating time-consuming repetitions.

We also observed that the "explicitness of the target" had a large impact. The results obtained for the explicit tasks were better than those observed in the implicit tasks. During the experiment, we noticed that this variable could be influenced by the subjects' knowledge. These were simple tasks and, if subjects did indeed possess any relevant knowledge, this took the form of the location at which the information might be found and the method required to access it. These subjects therefore performed more accurate searches than the others. The implicit tasks were more difficult to perform.

click to expand
Figure 20.2: Classification of the types of information-seeking tasks as a function of the recall and precision indices and of the effect of each of the variables

In the majority of cases, if subjects had any knowledge then this related to the content of the topic in question but not the location of the target. Furthermore, we observed that the subjects who had some knowledge concerning the content of the target topic also had a more accurate representation of the target than the others. Thus, as of the moment when they found the target they were in a better position to respond rapidly to the question since they simply had to check the information without needing to analyse it. However, in general the subjects were largely unsuccessful in this type of search. However, this might have been due to the experimental conditions which were not suitable for this type of task. The comments made by the subjects suggested that they needed more time either in order to perform a more extensive search or to summarise the results of the information they had found.

In situations such as those used in this protocol, subjects are working under optimum search conditions. They could not print the documents and their search was stopped if it took longer than 20 minutes. This suggests that the level of performance achieved in implicit searches is low when the search time is less than 20 minutes and the documents can only be consulted on screen. However, this does not exclude the possibility that such searches may be performed successfully under different conditions.

We observed that the "target distribution" and "target quantity" variables had a lesser effect. We can well imagine that this could be due to the choice of the modes associated with our variables. As we stated in the description of our results, we would probably have observed a greater effect for the "target quantity" variable if we had chosen different modes. We assume that this same observation applies to the "target distribution" variable. We observed that when the targets were distributed the subjects had difficulty finding them. When targets are distributed, subjects must continue their search until they have found all the information. The search for "distributed" information takes time and even though our subjects were not placed under any time constraints they expressed the need to find the information quickly. In many cases, the subjects abandoned their searches before finding all the relevant information. However, a "distributed" target may be spread over a greater or lesser number of pages. According to Tricot and co-workers [TRICOT, 1999], performance in a search for a "distributed" target can vary depending on the number of pages that exist within the search field. In other words, for a given number of pages that may contain the information, subjects will more easily find all the information if the targets are distributed through all the pages than if they are present on certain pages. We can therefore suppose that the size of the effect of the "target distribution" variable will vary as a function of the number of pages which may contain these targets or, in other words, the target distribution effect is undoubtedly very closely related to the effect of task "selectivity". Tricot and co-workers [TRICOT, 1999] have also hypothesised that one relevant characteristic of the task is the complexity of the procedure that has to be implemented, i.e. the number of different decisions that have to be made between the start and end of the activity. We should therefore also like to assess the effect of other objective variables such as the number of different decisions that have to be made in order to attain the goal (complexity of the procedure), but also whether the target is defined a priori or discovered a posteriori, the type of sensory modality (auditory/visual) involved in processing the target, the amount of data to be transferred or stored, the volume of data to be displayed (e.g. number of words).

The hierarchical organisation of the tasks allowed us to describe the types of information-seeking task more precisely. For each type of task, we have been able to evaluate the effect of each of the variables and the associated level of performance. Finally, we wish to test the hypothesis which holds that the more difficult a task is, the less suitable it is for a communications object which does not itself offer a high level of usability (for example: the more difficult an information-seeking task is, the less suitable it is for a mobile telephone). In a second stage, this should permit us to describe the set of tasks that can be performed using each type of communications object.

It is important to fix the limits of validity of our (present and future) results as well as of our approach. Our hierarchical organisation was developed within a precise context (limited time, information retrieval restricted to reading on screen) and the tasks that the subjects were asked to perform had the sole aim of enabling them to answer a question. However, Internet applications are more varied than this. For example, Bernstein [BERNSTEIN, 1993] distinguishes between three types of hypermedia application, namely information "mining", "manufacturing" and "farming". Information "mining" searches involve an attempt to extract information. In this type of task, the relevant information is a valuable resource which has to be extracted effectively and refined . Information "manufacturing" searches are searches which make it possible to design or draft a document. This type of information search conceives of the acquisition, refinement, assembly and maintenance of information as an ongoing endeavour . Information "farming" conceives of the " tending " of information as a continuous, co- operative activity conducted by groups of individuals working together in order to accomplish changing individual and communal objectives . Bernstein notes that the appraisal criteria used in these three activities are radically different and that any attempt to perform an activity in a system which has not been designed for this purpose is doomed to failure. Given Bernstein's categorisation, we may imagine that our hierarchical organisation is relevant for one defined type of application, namely information "mining" and, consequently, is unsuited for the evaluation of "manufacturing" or "farming" type tasks.




Communicating With Smart Objects(c) Developing Technology for Usable[... ]stems 2003
Linux Troubleshooting for System Administrators and Power Users
ISBN: N/A
EAN: 2147483647
Year: 2005
Pages: 191

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net