This section provides you with some ideas for how the QALL-ME Framework could be further improved and extended. If you just want to start using the framework or if you are already a happy user, then you can probably safely skip this section. Some of the following ideas may be rather simple to implement while others are really research topics that may not be solvable short-term.
Although it is not required by the QALL-ME Framework, the question pattern to query pattern
mappings (cf. 2.1.5) which are used in the QueryGenerator
implementations are currently probably
always manually created; at least this is the way all our demo
components work. It would be much nicer, though, if the question
patterns could somehow be learned automatically, maybe by deriving them
from question corpora and the used domain ontology.
The current implementation of the QAPlanner
uses reverse geocoding to find the country for
the geocoordinate location at which a question was posed. This country
information is then used to select the correct location specific
services. Currently the reverse geocoding parts are hard wired in the
QAPlanner
; it would be nicer, though, to have
this functionality in an extra web service so that one can easily switch
to another reverse geocoding implementation if necessary. Additionally
the fallback ways for finding the country information based on the
location string which is also given in each inquiry could be improved.
This fallback is always used in case the used reverse geocoding service
is unavailable.
The source code of the current implementation of the framework and of the demo components can always be improved further. Possibilities for making smaller improvements are often already indicated directly in the sources with “TODO” markers.
EntailmentTester
demo
componentsCurrently the performance of the EntailmentTester
components that are used in the different
language demo subsystems is far from the state-of-the-art in recognizing
textual entailment research. This situation could be improved by using
better textual entailment recognizers that are freely available, such as
the EDITS system.
Maybe it would be nice to use a UDDI registry in the QALL-ME Framework for the discovery and automatic binding of suitable web services. Currently all used web services in the framework have to be manually specified which is more or less a semi-dynamic binding only.
Currently the web services in the QALL-ME Framework are orchestrated with Java code. More flexibility and easy of use could be gained by using the open standard WS-BPEL for web service orchestration.
Further ideas for enhancements and feature wishes may be found in the project’s tracking system.