FalconAO: Aligning Ontologies with Falcon

Ningsheng Jian, Wei Hu, Gong Cheng, Yuzhong Qu

Introduction

Ontologies serve as the backbone of the Semantic Web, providing the structured vocabulary necessary for data interoperability. However, because ontologies are often developed independently by different organizations, they frequently vary in terminology and structure. This heterogeneity creates a significant barrier to data sharing. Falcon-AO is an automated system designed to address this challenge through robust ontology alignment, combining linguistic analysis with structural graph matching to find correspondences between disparate knowledge structures.

System Architecture

Falcon-AO is a hybrid alignment tool that integrates two primary matching components: a linguistic matcher (LMO) and a structural matcher (GMO). The system follows a multi-stage process where ontologies are first parsed and then processed by these matchers. The final alignment is produced by an integration module that dynamically weights the outputs based on the comparability of the input ontologies.

The Falcon-AO system architecture, illustrating the flow from ontology parsing to alignment integration.

Linguistic Matching (LMO)

The Linguistic Matching for Ontologies (LMO) component calculates similarity using two distinct techniques:

  1. Lexical Comparison: This method employs "edit distance" to measure the character-level similarity between entity names.
  2. Statistical Analysis: Utilizing a Vector Space Model (VSM), LMO treats each entity as a "virtual document." This document includes terms from the entity’s name, labels, comments, and the names of its neighboring entities. By applying TF-IDF weighting, the system can capture the semantic context of an entity even when its primary name is ambiguous.

The final linguistic score is calculated as a weighted combination: $0.8 \times \text{Statistical} + 0.2 \times \text{Lexical}$.

Structural Matching (GMO)

The Graph Matching for Ontologies (GMO) component represents ontologies as directed bipartite graphs. It operates on the intuition that two entities are likely similar if they participate in similar "triples" (subject-predicate-object relationships). A key innovation in Falcon-AO is the use of LMO results as "external mappings" to anchor the structural search. This linguistic-first approach significantly reduces the computational search space for the graph matcher.

Integration Strategy

To merge the results from LMO and GMO, Falcon-AO calculates two meta-metrics: Linguistic Comparability (LC) and Structural Comparability (SC). These metrics assess the overall similarity of the two ontologies. If the ontologies exhibit low linguistic comparability (e.g., they use different languages or naming conventions), the system automatically places greater weight on the structural matches identified by GMO.

Experimental Results

The system was evaluated using the OAEI 2005 (Ontology Alignment Evaluation Initiative) benchmarks, demonstrating high precision and efficiency across various test scenarios.

Performance Highlights

  • Near-Perfect Alignment: On benchmarks with high linguistic similarity (Tests 101–104 and 221–247), Falcon-AO achieved an F-Measure between 0.99 and 1.0.
  • Robustness to Noise: The system proved highly resilient to naming noise. In tests where local names were replaced with random strings (Tests 201–210), Falcon-AO maintained an average F-Measure of 0.95 by leveraging labels, comments, and structural context.
Average performance metrics for Tests 201-210, demonstrating robustness against randomized local names.
  • Efficiency: While simple alignments were completed in under 5 seconds, more complex structural tasks (such as Tests 201–210) required approximately 63 seconds on average.

Conclusion and Future Work

Falcon-AO represents a significant step forward in automated ontology alignment by intelligently combining linguistic and structural evidence. While highly effective, the current version has limitations, such as the lack of a background lexicon (like WordNet) for synonym handling and a restriction to one-to-one mapping cardinalities. Future developments aim to incorporate more sophisticated comparability metrics and improve the scalability of the GMO component for large-scale medical ontologies.

Summary updated

Abstract

Falcon-AO is an automatic tool for aligning ontologies. There are two matchers integrated in Falcon-AO: one is a matcher based on linguistic matching for ontologies, called LMO; the other is a matcher based on graph matching for ontologies, called GMO. In Falcon-AO, GMO takes the alignments generated by LMO as exter-nal input and outputs additional alignments. Reliable alignments are gained through LMO as well as GMO according to the concept of reliability. The reliabil-ity is obtained by observing the linguistic comparability and structural comparability of the two ontologies be-ing compared. We have performed Falcon-AO on tests provided by OAEI 2005 campaign and got some prelim-inary results. In this paper, we describe the architec-ture and techniques of Falcon-AO in brief and present our results in more details. Finally, comments about test cases and lessons learnt from the campaign will be presented.
BibTeX
@inproceedings{jian2005falconao,
  author = {Ningsheng Jian and Wei Hu and Gong Cheng and Yuzhong Qu},
  title = {{FalconAO: Aligning Ontologies with Falcon}},
  booktitle = {Integrating Ontologies},
  year = {2005},
  url = {https://openreview.net/forum?id=9u5E1FpRmh6},
}
Figure