Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semantic Alignment of Multilingual Knowledge Graphs via Contextualized Vector Projections

Abhishek Kumar

TL;DR

The paper addresses cross-lingual knowledge graph alignment by enriching multilingual entity descriptions through template-based verbalization and using fine-tuned multilingual transformers to produce contextual embeddings. It combines cosine similarity with a Hungarian algorithm-based thresholded matching, augmented by HermiT reasoning to enforce semantic constraints, achieving strong recall and competitive F1 on the OAEI-2022 MultiFarm benchmark with a notable 16% improvement over baselines. Key contributions include a novel verbalization-driven alignment framework, comprehensive ablations demonstrating verbalization’s impact, and robust evaluation across multiple language pairs using efficient nearest-neighbor search. The approach is lightweight and privacy-friendly, suitable for domains where data cannot be moved to external LLMs, while offering practical pathways for future enhancements like contrastive learning and XAI integration.

Abstract

The paper presents our work on cross-lingual ontology alignment system which uses embedding based cosine similarity matching. The ontology entities are made contextually richer by creating descriptions using novel techniques. We use a fine-tuned transformer based multilingual model for generating better embeddings. We use cosine similarity to find positive ontology entities pairs and then apply threshold filtering to retain only highly similar entities. We have evaluated our work on OAEI-2022 multifarm track. We achieve 71% F1 score (78% recall and 65% precision) on the evaluation dataset, 16% increase from best baseline score. This suggests that our proposed alignment pipeline is able to capture the subtle cross-lingual similarities.

Semantic Alignment of Multilingual Knowledge Graphs via Contextualized Vector Projections

TL;DR

The paper addresses cross-lingual knowledge graph alignment by enriching multilingual entity descriptions through template-based verbalization and using fine-tuned multilingual transformers to produce contextual embeddings. It combines cosine similarity with a Hungarian algorithm-based thresholded matching, augmented by HermiT reasoning to enforce semantic constraints, achieving strong recall and competitive F1 on the OAEI-2022 MultiFarm benchmark with a notable 16% improvement over baselines. Key contributions include a novel verbalization-driven alignment framework, comprehensive ablations demonstrating verbalization’s impact, and robust evaluation across multiple language pairs using efficient nearest-neighbor search. The approach is lightweight and privacy-friendly, suitable for domains where data cannot be moved to external LLMs, while offering practical pathways for future enhancements like contrastive learning and XAI integration.

Abstract

The paper presents our work on cross-lingual ontology alignment system which uses embedding based cosine similarity matching. The ontology entities are made contextually richer by creating descriptions using novel techniques. We use a fine-tuned transformer based multilingual model for generating better embeddings. We use cosine similarity to find positive ontology entities pairs and then apply threshold filtering to retain only highly similar entities. We have evaluated our work on OAEI-2022 multifarm track. We achieve 71% F1 score (78% recall and 65% precision) on the evaluation dataset, 16% increase from best baseline score. This suggests that our proposed alignment pipeline is able to capture the subtle cross-lingual similarities.
Paper Structure (21 sections, 1 equation, 5 figures, 4 tables)

This paper contains 21 sections, 1 equation, 5 figures, 4 tables.

Table of Contents

  1. Introduction and Motivation
  2. Technical Novelty and Differentiation
  3. Key Contributions
  4. Related Work
  5. Embedding-Based Alignment Architecture
  6. Extended Baseline Comparison
  7. Error Analysis and Failure Patterns
  8. Comprehensive Ablation Study
  9. Embedding Space Visualization
  10. Multilingual Semantic Verbalization Strategies
  11. Evaluation and Results
  12. Datasets and Experimental Setup
  13. Results Summary
  14. Interpretation of Results
  15. Ablation Study
  16. ...and 6 more sections

Figures (5)

  • Figure 1: Semantic alignment pipeline using contextual embeddings from multilingual transformer models.
  • Figure 2: Distribution of error types in cross-lingual alignment
  • Figure 3: t-SNE visualization of cross-lingual entity embeddings
  • Figure 4: Cross-lingual alignment pipeline using vector similarity and matching heuristics.
  • Figure 5: Ablation F1 impact.