In this paper an application is presented that allows the construction of conjunctive queries (represented as labeled graphs), the presentation of the results of the queries (as tables), and the selection of options for specialization or generalization of the graph queries. The graph labels correspond to the query vocabulary, which is obtained by conceptual scaling of the database. Each scale provides the specialization and generalization options for a particular kind of graph node. At any time, the user is shown the current query, the result table, and a set of options to modify the query.
The tool is realized as a web application, using Python 3 server-side. Supported browsers are current versions of Firefox and Chrome, which support modern standards like HTML5 and Javascript ES6. The Granada package contains a Copyright c 2019 for this paper by its authors. Copying permitted for private and academic purposes. le app.wsgi, which is the interface to the web server (cf. https://www.python. org/dev/peps/pep-0333/). For simplicity, the le can also be run as a script with python3 app.wsgi, which starts the app in Python's built-in web server. The initial application screen (Fig. 1) can then be found at localhost:8081. The navigate and edit pane (left and middle in Fig. 1) list all databases known to the system (both lists are identical). Initially, the lists will be empty. To include a database in the list, the relevant information (user, password, host and database name) is entered in the create section (right section in Fig. 1). This creates a JSON le on the server with connection information and additional information obtained from the database (e.g. tables and columns). We refer to this le (and the information it contains) as a binding ; a name for the binding is optionally speci ed on creation (this defaults to the database name). Once a binding is created, it is listed in both the navigate and edit panes.
Clicking on an item in the edit pane (middle in Fig. 1) displays the associated
binding in the edit view (Fig. 2). The tables in the database are listed on the
left of Fig. 2, and we refer to them as sorts. To the right of the sort list, all
binding information related to a particular sort (selected in the list) can be seen.
The cyan entries in the properties pane are the table columns, and each is listed
with its SQL datatype (similar to how attributes and methods are listed in an
API documentation). Green entries in the relations pane represent foreign keys
on or into the respective table/sort. Foreign keys are only included in the initial
binding if they are explicitly speci ed in the database. Otherwise, they have
to be speci ed manually, using the form at the bottom of the relations pane.
We refer to the entries in either the properties or relations section as
manyvalued attributes. Such attributes can be conceived as unary functions (a foreign
key is conceived as a Boolean function on object pairs, i.e. as the characteristic
function of the underlying relation). The output pane (top of Figure 2) speci es
how a table row is printed as text (by default, the primary key is printed). The
drop-down menu on the right-hand side of each attribute is used to con gure
whether the attribute can be used in queries and, if yes, what search mask is
used to specify conditions (in FCA terms, the drop-down menu is used to scale
the attribute, cf. [
The navigation view (cf. Fig. 3) is opened when the user clicks on a binding in the navigate section in Fig. 1. We assume here that the database has already been con gured for navigation, either in the editor view as described above, or automatically by the system, using sensible defaults.
The navigation view is partitioned into four panes, showing the query graph (top right), the result table (lower right) and associated options (left windows). The initial query graph consists of only a single rectangular node, labeled Any:x1. It describes the set of all objects in the database (i.e. all rows of all tables). Only in this particular case is an associated result table not shown in the lower right pane, this is because there is no corresponding SQL query for the query graph.
The only action that is initially available to the user is the restriction of x1 to one of the sorts (i.e. table names) available in the database. The `Literature' database, which is used in the examples, consists of two relations/tables Author and Book. These can be selected in the label view (upper left). The numbers next to the sorts represent the number of rows in the respective tables. Selecting one of the sorts (cf. Fig. 3) updates the query graph (the node label now shows the selected sort, see top right), produces a table which lists all objects of that sort (bottom right), and shows all properties and relations associated with the objects of that sort in the link view (bottom left). Unchecking the sort in the label view leads back to the previous query. This allows one to quickly examine the available sorts, their associated properties and relations. What properties and relations are exactly shown can be speci ed in the bindings editor (Fig. 2).
The Author sort has two properties, DOB (date-of-birth) and nationality,
listed in the link view under properties. The eye icon is an a ordance that allows
one to view the respective property in the result table (by adding an additional
column in each case). Clicking the eye icon again toggles the view on the
respective column. This does not modify the graph. Clicking the plus control symbol
on the other hand, adds a relation node to the graph (relation nodes are drawn
as rounded nodes), which represents a condition on the many-valued attribute.
Some initial condition is given, which may or may not restrict the current result
set. In the case of the DOB attribute, the initial condition speci es that authors
are born between 1800 and 2000 (a suitable range is computed which does not
restrict the result set). A minus now appears in place of the plus control symbol;
clicking on it destroys the relation node (and thus removes the condition).
Clicking the button with the property name shows options to modify the condition
in the label view (top left), if the property is active (i.e. an associated relation
node exists in the graph). For the DOB property, the label options are given by
a slider with two handles, which allows one to select an interval of permissible
values (cf. Fig. 4). The nationality attribute is associated with a di erent kind
of scale, which allows the user to specify that the string-valued attribute must
begin with a certain pre x. A pre x can be entered in a text eld in the label
view, or chosen out of a list of occurring values (cf. Fig. 5). The plus symbol
on a relation link creates a relation node together with object nodes for the
associated objects, with a corresponding sort. For example, clicking on the plus
symbol for the wrote relation creates the corresponding relation node plus an
associated Book node (cf. Fig. 6). Clicking on the large button with the relation
name on it shows the pairs in the relation that satisfy the query constraints. If
a relation between objects has an associated value (an example from [
Granada is based on a navigation principle previously described in [
Sparklis [