Currently the link between the Machine Discovery Component and the HyperSpace Visualiser is effectively one way. Although features in the visualisation can be selected, and the data included in the features identified, there is no simple way to feed this information back to the MDC, so that the features can be quantified into formally specified discoveries. Adding this feature to the system would enable an iterative discovery process.
A number of link weighting and pruning algorithms have been tried. Future work will include the evaluation of other algorithms, particularly those based on ranking of links and the allocation of a fixed amount of strength to a node, which is then shared out amongst its links.
Another development is to use a single prototype can be used to represent multiple items which would be treated in the same, or very similar manner in the visualisation.
Prototypes for clusters can be formed by generating expressions which capture the similarities between the data items in those clusters, or alternatively by selecting the data nodes with most links. These prototype can serve two purposes. Firstly, prototypes are also useful discoveries in their own right. Secondly, replacing a number of nodes in a cluster with a single prototypical node leads to a reduction in visual clutter an increase in rendering speed and thus allows the system to scale to cover larger amounts of data.
The visualisation currently scales well to cope with larger amounts of data. For a large database, the obscuring of nodes by other nodes could be a problem. Prototypes and transparency can help ease this.