Ontology-Based Depth–Breadth Matrix

• Depth–Breadth Matrix (Ontology-Based) is a framework that quantifies hierarchical taxonomies by measuring maximum tree depth and category breadth using ontological methods.
• It utilizes local ontologies and semantic distance thresholds to dynamically decide when to merge, add a sibling, or insert a child category.
• The approach enables algorithmic evaluation and reconfiguration of directory structures to maintain balance and improve semantic organization.

A Depth–Breadth Matrix in the context of ontology-based administration of Web directories formalizes the structural organization of hierarchical taxonomies by quantifying both their depth (the longest root-to-leaf path) and breadth (the number of categories at each level). This abstraction provides a framework for analyzing, comparing, and algorithmically evolving directory structures using semantic criteria derived from ontologies. While the term "Depth–Breadth Matrix" does not explicitly appear in the foundational work by Horvat et al., the essential mathematical and algorithmic components for constructing such a representation underlie their approach to Web directory management and automation (Horvat et al., 2013).

1. Hierarchical Web Directory Structure

A Web directory $D$ is defined as a rooted tree of categories $C$, where each category $c \in C$ occupies a level $l(c) \in \{1,2,\dots,L\}$, with the root always at $l=1$. The depth $L$ is formally the maximal level:

$L = \max_{c \in C} l(c)\,.$

Breadth at a given level $i$ is quantified as the number of categories present at that level:

$$C_i = \{\,c\in C\mid l(c)=i\,\}, \qquad b(i) = | C_i |\,.$$

Locally, a per-node branching factor is given by the number of children of a category $c$:

$$\mathrm{bf}(c) = | \{ d \in C \mid d \ \text{is a child of } c \} |\,.$$

This structure serves as the backbone for semantic organization and evaluation using ontological similarity metrics.

2. Ontology-Driven Semantics for Categorization

Each directory category $c$ possesses a local ontology $O_c$ generated via a semantic-extraction function $\sem\colon R \to O$ applied either to the set of resources contained in $c$ or recursively constructed from the ontologies of its child categories. This semantic enrichment enables automated, meaning-preserving directory construction and refinement, as sub-categories and mergers are not solely determined by syntactic or heuristic mechanisms but by quantitative semantic similarity (Horvat et al., 2013).

3. Ontological Distance and Decision Criteria

Dynamic construction and maintenance of the directory’s structure is governed by the ontological distance between categories:

$\dist(O_a, O_b) = \frac{1}{\sim(O_a, O_b)}$

where $\sim\colon O \times O \to [0,1]$ denotes the semantic similarity function. Two primary thresholds regulate when new categories emerge and at which position:

• Minimal horizontal semantic distance ($\mathit{mindist}_H$): Threshold for adding siblings at the same level.
• Minimal vertical semantic distance ($\mathit{mindist}_V$): Threshold for adding children, increasing the depth.

A resource ontology $O_{\mathit{new}}$ is:

• Merged if $\dist \le \min\{\mathit{mindist}_H, \mathit{mindist}_V\}$,
• Inserted as a sibling if $\mathit{mindist}_H < \dist \le \mathit{mindist}_V$,
• Inserted as a child category (increasing depth) if $\dist > \mathit{mindist}_V$.

These thresholds render the depth and breadth dynamics responsive to semantic rather than purely structure-driven considerations.

4. Formal Representation: Depth–Breadth Matrix

Although the explicit terminology "Depth–Breadth Matrix" is not used, the following representation is directly constructed from the paper's definitions:

• For directory depth $L$ and breadths $b(i)$ at each level:

$\mathbf b = [ b(1)\ b(2)\ \dots\ b(L) ]$

Or as a column vector:

$\mathbf b^T = [b(1),\ b(2),\ \dots, b(L)]^T$

This vector suffices to profile a directory’s level widths, facilitating comparative analysis across directories, and tracking evolutionary changes as resources are added or hierarchical reconfigurations occur.

Level ($i$)	Number of Categories ($b(i)$)
1	$b(1)$
$\vdots$	$\vdots$
$L$	$b(L)$

A full matrix embedding both breadth (per level) and depth (level index) can further be constructed, but for single-directory analysis, the breadth vector captures all essential structure information.

5. Ontology-Based Directory Construction Algorithm

The construction process automates hierarchy formation via iterative evaluation against semantic distance thresholds. The pseudocode based on Horvat et al. is as follows:

for each resource r in R: Onew = sem(r) if only root exists: create new category at level 2 with ontology Onew place r in new category else: c_star = argmin_c dist(Onew, O_c) d = dist(Onew, O_{c_star}) if d > mindistV: create Cnew as child of c_star, level = l(c_star)+1 place r in Cnew elif d > mindistH: create Cnew as sibling of c_star, level = l(c_star) place r in Cnew else: merge r into c_star O_{c_star} = O_{c_star} + Onew # unify ontologies

Increasing depth and breadth are direct consequences of when and where new categories are spawned. Each action updates the depth $L$ and the breadth vector $\mathbf b$ accordingly.

6. Application: Directory Analysis and Quality Evaluation

The depth $L$ and breadth vector $\mathbf b$ function as objective, quantifiable measures of a directory’s structure. Algorithmic procedures can automatically reconfigure the hierarchy to optimize for a desirable “Depth–Breadth Matrix”—for instance, balancing the tree or enforcing targets for maximum depth or even-level widths. These measurements serve as a basis for structural quality evaluation, longitudinal monitoring, and feedback in the ongoing development or realignment of Web directories (Horvat et al., 2013).

A sample scenario demonstrates these processes:

• Initial root, $L = 1$.
• Inserting $r_1$: directory becomes $L = 2$, $\mathbf b = [1, 1]$.
• Inserting $r_2$ (depending on $\dist$ and thresholds): possible outcomes $L = 3$, $\mathbf b = [1, 1, 1]$ (new child), or $L = 2$, $\mathbf b = [1, 2]$ (new sibling), or a merge if ontologies are highly similar.

7. Significance and Extensions

The ontology-based approach enables the Depth–Breadth Matrix (or vector) to not only document structural properties but also connect directly to semantic considerations. The semantic thresholds $\mathit{mindist}_H$ and $\mathit{mindist}_V$ introduce a principled mechanism for structure evolution closely tied to the meanings and relationships within the knowledge domain. A plausible implication is that such ontologically-driven matrices could guide the automated optimization or refactoring of directory structures towards arbitrarily specified depth–breadth profiles, supporting a broad class of use cases in knowledge organization and retrieval system design (Horvat et al., 2013).