5. AIS use in practice
The preceding two sections have given a formal
description of the components of an adaptive information
system (AIS) and a compatible, adaptive model of data. In
this section we address a number of practical concerns
that arise when AIS are implemented and used. We
describe the use of an AIS in practice to meet the needs
of a particular organisation.
5.1. Impact of multiple concept versions
In an AIS, as in document and graph databases, each
item of data is stored with its own structure, which
reflects the version of the concept used when the data
item was last saved. Therefore, data relating to conflicting
versions of each concept can coexist. This behaviour
is very different from that of conventional systems,
which are rendered completely or partially unusable
following schema change, until modified to match the
new schema. An AIS will continue to function after
schema change, and the user can work with the system
as usual. There is no need for software maintenance,
nor any immediate need for data transformation; the
data may be upgraded to the new structure gradually, or
not at all.
Many questions remain about how AIS will operate in
practice. For example, it is unclear precisely what form
database queries should take; this is a topic for further
research (see Section 6). Because multiple concept definitions
may be present, an AIS must select the correct
S. McGinnes, E. Kapros / Information Systems 47 (2015) 33–50 45
definition to use for each data instance. Typically, the latest
concept definitions are used during data entry, to control
user interface behaviour and to validate newly-recorded or
modified data. But, for data retrieval and display, the
retrieved data may conform to earlier versions, and therefore
all of the definitions in the relevant concept family
must be reconciled, at run time.
To illustrate this process, Fig. 12 shows a sample XML
data file fragment. The customer entries conform to
different concept definitions; they have conflicting lists
of attributes. The AIS merges the definitions to produce the
domain superset, presenting a unified view of the concepts
in the concept family.
The ability to accommodate multiple concept versions
also raises issues of data integrity [31]. A detailed treatment
is beyond the scope of this paper, but we note that
a limited form of data integrity is possible. Each item of
data in the data file conforms to a particular concept
definition, and therefore was valid when last saved.
Referential integrity cannot be violated since data may
be deleted only through the AIS, which checks for
dependencies. Relationships will therefore remain traversable.
The key difference from relational databases is
that, while the concept definition applicable to any item
of data can be retrieved, all data for a particular concept
family cannot be assumed to hold to any particular
concept definition.