Chapter 9 Conclusions and Future Work
9.3 Future Work
9.3.1 Research Issues
• Classification of Links
An extension to this work would be to look at how qualitatively the links in each dimension can be categorised into one dimension and not in another category. For example, the links in the Expertise MDL were categorised into three dimensions, i.e.
Subject, Language, and Assessment dimensions, and the Subject links applied the “Transformation Model” into their subclassification, which resulted in Raw Material,
Operations, and Output. Each of these sub-dimensions provides four options for the user to choose, i.e. ‘no links’, ‘basic links’, ‘advanced links’, ‘all links’, for their presentation of links. Based on the current prototype, the ‘rule of thumb’ has been employed to classify the ‘basic’ and ‘advanced’ links. Therefore, in order to serve good quality links, the domain expert might still be required for the purpose of links categorisation.
• The Annotation of Links
One of the recommendations in the experiment was to differentiate links from different sub-dimensions, even though the use of different colours was already adopted to simply distinguish the Expertise links from the other two links (Inquiry and Glossary). To strengthen the MDL concept and its system implementation, the system developed should allow users toknow from which dimension in the MDLs the links are appearing, so that the users might be able to configure the system for the best personalisation of links to suit their expertise and preference. The possible techniques that might serve this purpose include a number of link annotation techniques such as colouring links from different dimensions distinctively; the concept as applied in ‘Fluid links’ (Zellweger et al., 1998) i.e. annotating links with additional information about the destination page, whereby the annotations are added below a link anchor after the user moves the mouse over it (Tsandilas and schraefel, 2003); adapting the font size of the links (Tsandilas and schraefel, 2003); and using small icons to symbolise link types (Weinreich and Lamersdorf, 2000).
In addition, Weinreich and Lamersdorf (2000) provided a short survey of methods to present link types, which can be firmly implemented with the MDL concept, for instance, link colour, mouse pointer (i.e. the pointer changes according to the link type when it hovers over a link), and popups or rollovers (i.e. small floating windows appearing next to the mouse cursor when moving over a link).
• Advanced Text Processing or Information Retrieval Techniques
Other research issues raised in the experiment involve looking into more advanced text processing and parsing techniques. First, some users pointed out that they did not like seeing repeated words. In interviews with the users after the
evaluation, it transpires that they thought it would be more useful and less distracting if they could be provided with the option asking them if they wanted to see the same keyword presented and appeared as links more than once. For instance, the word ‘gluten’ coming from the Subject dimension and the Output sub-dimension, would appear as links everytime the user chooses the option ‘basic links’ for the Output sub- dimension. As a matter of fact, every paragraph containing the word ‘gluten’ would become a link to its additional explanation. However, when the user selects the option ‘advanced links’ for the same sub-dimension, the word ‘gluten’ would no longer be a link. Nevertheless, their concern was that they wished to only see the word ‘gluten’ as a link once, or at least they would have an option if they wanted to see the same word as links more than once or not.
• Adaptability to Adaptivity
Lastly, a few users pointed out that it might be cumbersome in a bigger hyperspace if there were too many dimensions and sub-dimensions for them to make selection for the presentation of links. This could dissuade the users from configuring and making use of the system. As a consequence, the shift from ‘personalisation or adaptability’ to ‘adaptivity’ might discontinue the argument. That is, instead of replacing the previous user selection on link presentation with a new selection, as now, these past records could possibly be logged to refine the degree of accuracy for providing satisfactory corresponding links in relation to each expertise dimension. The system could automatically apply this knowledge to provide augmented links. In other words, some kind of inference engine or mechanism is required.
One possible approach might be to use an information retrieval technique, such as
Term Frequency Inverse Document Frequency (TFIDF), to calculate the frequency of the selected option of each expertise dimension by requesting the users perform some tasks and set this weight against the statistical frequency of the options occuring in a random set of sample documents (Joachims et al., 1997).
However, we need to be sure that the adaptivity provided by the system will not overshallow the ‘free-form’ approach provided by the OH research. As Conlan (2003) also pinpointed, the balancing control made available to users and the transparency of the adaptivity need to be considered, as these issues can introduce design arguments.