Previous Up Next

4  Browsing

The main function of Camelis is to browse a context. The specificity of Camelis browsing is to allow for an arbitrary combination of querying and navigation. Every query defines a navigation place, and reciprocally every navigation place is defined by a query. A query also defines an extent that is the set of objects whose description is subsumed by the query. So a navigation place can always be seen at the logic-level through the query, and at the object-level through the extent. The combination of a query and its extent is called a concept.

At any time, Camelis displays a coherent concept view that is the combination of a query, a set of objects, and a set of features. The set of objects must be the extent of the query. The set of features is structured into a taxonomy according to the subsumption ordering. By default, only features occuring frequently in the extent are displayed, so that the taxonomy of features is dynamically adapted to the current concept view. The graphical interface reflects this concept view: As a consequence of the equivalence between queries and navigation places, navigation links are query transformations, and query transformations are navigation links. Navigation links are defined as the composition of a selection and a navigation mode. A selection can be a set of features or a set of objects in the concept view. When several features are selected, they form a disjunction. A feature selection can be modified by check and radio buttons at the top of the navigation window: add of negation in front of the selection (NOT), or replacement of equalities by inequalities (=/≥/≤). There are 3 navigation modes. Zoom and pivot apply to feature selections, while querying by examples applies to object selections. Moreover, as in web browsers it is possible to go back and forward in the history of visited queries.

4.1  The Navigation Window and Query Area

The navigation window displays a set of trees of features as tabs that initially contains a single tree, whose root is the most general feature and that is completely collapsed, so that only most general features are visible. More specific features can be made visible by expanding any element in any tree. For ease of navigation between different views, a feature can be expanded as the root of a new tab by activating the command Expand as tab in the contextual menu of the navigation window. This new tab can be seen as a zoom on the full feature tree.

Note that a same feature may appear more than once in the trees. This is because the set of features is in fact partially ordered by subsumption and is not a hierarchy. A feature can be inserted into a tree through the command Logic > Add Feature.... Where it is inserted is automatically determined by the subsumption relation. A set of selected features can be hidden through the command Logic > Delete Features. A feature can be replaced by another feature in the whole context by activating the command Replace..., in the contextual menu, on the feature to be replaced. These 3 commands enable users to customize the navigation window contents according to their needs.

The number on the left of each feature is the number of objects in the current extent that is subsumed by the feature, i.e. the count of the feature. The sum of these numbers is usually greater than the size of the extent because objects are generally subsumed by many features. There lies a big difference with hierarchical systems where objects belong to only one path in the tree. This means that, for example, a user doesn't have to choose a song either by its artist or by its genre, but can successively select an artist and a genre in any order. Furthermore, after selecting an artist, only relevant genres are suggested, and reciprocally, after selecting a genre, only relevant artists are suggested. The font color of features depends on their count and their relation with the query: The font scale of features and their count is proportional to the ratio between the logarithm of the count and the logarithm of the size of the extent, i.e., the number of objects in the extent. At the left of the count, a background grey-level indicates the ratio between the logarithm of the count and the logarithm of the total count of the feature in the context: white means 100%, and dark means 0%. This total count of features can be made visible over the background grey-level by activating an option in the menu Browsing. If 2 features cover the same objects in the extent, a same color is used as the background of their count (which are necessarily equal). Font color, font scale, background grey-level, background count color are activated by default, but can be desactivated in the menu Browsing.

The initial query is all so that the initial extent is the set of all objects of the context. Then this query can be modified by editing it in the query area, or by following navigation links. The selection of a navigation link consists in defining a selection, and then choosing a navigation mode. Querying by examples applies to an object selection, and defines the new query as the conjunction of all features that are shared by all selected objects. Pivot applies to a feature selection, and replaces the query by the selection, hence starting a new search from the features of the current concept view. Zoom applies to a feature selection, and either specializes (zoom-in) or generalizes (zoom-out) the current query. If the selection covers all objects in the current extent, a zoom-out is performed, and the query is generalized, hence a larger extent. Otherwise, a zoom-in is performed, and the query is specialized, hence a smaller extent. The transformation of a query by zoom is defined as follows. The current query is split into a conjunction of simpler parts. If the selection is a part of the query, it is removed from the query. Otherwise, every part that subsumes or is subsumed by the selection is removed, and the selection itself is added to form the new query.

The following is a navigation scenario, alternating queries and navigation links: Double-clicking on a feature is a shorthand for zooming on a feature selection made of this single feature. Double-clicking on an object is a shorthand for querying by examples on an object selection made of this single object.

Each time a new query is defined, both navigation and extent windows are refreshed, and it is added to the history of visited queries. Users can move backward and forward like in Web browsers through commands Back and Forward (available in the browsing menu, and as buttons). The additional command Refresh enables the refreshment of both navigation and extent windows when necessary. The button Home leads directly to the most general query, all.

In order to accelerate the browsing of a context, there are preselection modes that can be activated from the menu Browsing. With Object preselection, whenever a set of features is selected in the navigation window, the extent that would result from a zoom on this selection is highlighted as a subset of the current extension. This makes it possible to quickly see for any feature, which objects in the extent belong to it. With Feature preselection, whenever a set of objects is selected in the extent window, each feature in the navigation window gets a background grey-level that is proportional to the count of this feature in the object selection out of the count in the current extent: white means 100%, and dark means 0%. This makes it possible to quickly see for any set of objects, which properties they have, and how frequent they are in the object selection. When a single object is selected, it makes it possible to see the properties of this object (background is white).

When expanding a feature in order to see more specific features, default options are used. These options can be customized on each feature by the command Expand... in the contextual menu. We now describe these options and give their default value. Expanding options defined on some feature are inherited by sub-features. Expanding options of the root (invisible) can be defined though the command Browsing > Expand....

In addition to inserting and hidding features, users can customize the tree of features by expressing axioms. An axiom is specified by a pair of features (f,g), so that the first become subsumed by the second. The support for axioms depends on the chosen logic, and expressing an axiom may have no effect. An axiom (f,g) cannot be removed directly, but only by stating an axiom (f,h) where h subsumes g. The most common use of axioms is for building a taxonomy of terms. The first way to state an axiom is by a copy and paste mechanism available from both the contextual menu of the navigation window, and the menu Logic: to state an axiom (f,g) first select and copy feature f, then select feature g, and paste. In order to state all axioms between 2 sets of features, simply selects several features where relevant. When no feature is selected when pasting, features from the current query are used instead. This is useful when the subsuming feature is the root feature all as it is not visible in the navigation window. The second way to state an axiom is simply to drag and drop a feature on another feature.

Unlike hierarchies that are built top-down, taxonomies in LIS are rather built bottom-up, i.e. from more specific to more general features, because objects are described first, and classes emerges from these descriptions. So a common operation is to join a set of features under a more general feature. This is precisely what is provided by the command Join... in the contextual and logic menus: first select the specific terms, then apply the command Join..., which asks for a general joining feature.

4.2  The Extent Window

The extent window displays a partial list of objects in the current extent. The reason for displaying only a partial list is that displaying of very large extents is time-consuming and of no practical use. The size of the current extent is visible at the top right of the extent window, and the range of the displayed objects at the top left. The arrows on both sides of these figures provides page-scrolling in the extent: respectively from left to right, first page, next page, previous page, last page.

The extent is split in two tabs. The tab Picto contains pictures and photos, which are displayed by a thumbnail and a name. The tab Texto contains other objects, which are displayed by a short text. A contextual menu is accessible from the extent window with operations applying to selected objects. The first commands are possible actions on selected objects, and are based on their action arguments (see Section 7). The last command provides options on the extent window such as: The one but last command of the contextual menu, Intent, is an additional navigation mechanism that allows to query a context from examples. Given a set of selected objects this command sets the current query to the conjunction of features that are common to all selected objects. When applied on only one object, this can be performed by double-clicking on it, and makes visible all its descriptors in the navigation window.

Other commands are about updating objects, and are described in Section 6.

Previous Up Next