Multilingual Question Answering over Knowledge Graphs / Aleksandr Perevalov ; 1. Reviewer Prof. Dr. Axel-Cyrille Ngonga Ngomo, 2. Reviewer Prof. Dr. Andreas Both, 3. Reviewer Prof. Dr. Mike Scherfner. Paderborn, 2025
Inhalt
- Titlepage
- Abstract
- Kurzfassung
- Publications
- Acknowledgement
- Contents
- 1 Introduction
- 1.1 Motivation and Problem Statement
- 1.2 Research Question and Considered Research Gaps
- 1.3 Thesis Structure
- 2 Preliminaries
- 2.1 Natural Language Processing
- 2.1.1 Language Modeling
- 2.1.2 Sequence Labeling
- 2.1.3 Text Classification
- 2.1.4 Sequence-to-Sequence Conversion
- 2.1.5 Vector Semantics and Embeddings
- 2.1.6 Multilinguality and Low-Resource Languages
- 2.1.7 Large Language Model-based Agents
- 2.2 Semantic Web and Knowledge Graphs
- 2.3 Question Answering and Information Retrieval
- 3 Systematic Survey on Multilingual Knowledge Graph Question Answering
- 3.1 Introduction
- 3.2 Systematic Review Methodology
- 3.2.1 Selection of Sources
- 3.2.2 Initial Publications Selection
- 3.2.3 Extraction and Systematization of the Information
- 3.3 Systematic Surveys about Question Answering over Knowledge Graphs
- 3.4 Multilingual Question Answering Systems over Knowledge Graphs
- 3.4.1 Review of the Selected Systems
- 3.4.2 Evaluation Results of the Reviewed Systems
- 3.4.3 Language Coverage by the mKGQA Systems
- 3.4.4 Summary
- 3.5 Benchmarking Datasets for Multilingual Question Answering over Knowledge Graphs
- 3.6 Discussion and Challenges
- 3.7 Conclusion
- 4 From QALD-9 to QALD-9-plus: A Novel Benchmark for mKGQA
- 4.1 Introduction
- 4.2 Qualitative Analysis of the Original QALD-9 Benchmarking Dataset
- 4.3 An Approach to Creating Translations of Questions via Crowdwsourcing
- 4.4 Migration from DBpedia to Wikidata Knowledge Graph
- 4.5 QALD-9-plus Dataset Statistics Overview
- 4.6 Quantitative and Qualitative Question Analysis
- 4.7 Discussion
- 4.8 Conclusion
- 5 Machine Translation as an Alternative for mKGQA
- 5.1 Introduction
- 5.2 Machine Translation Approach for KGQA Systems
- 5.3 Experimental Setup
- 5.4 Evaluation and Analysis
- 5.4.1 Evaluation Results for Original Questions
- 5.4.2 Machine Translation Quality
- 5.4.3 Evaluation Results for Machine Translated Questions
- 5.5 Discussion
- 5.6 Conclusion
- 6 Improving mKGQA Systems' Quality by Verbalizing and Filtering Incorrect Output
- 6.1 Introduction
- 6.2 The SPARQL Query Filtering Approach
- 6.3 Experimental Setup
- 6.4 Evaluation and Analysis
- 6.5 Discussion
- 6.6 Conclusion
- 7 Large Language Model-driven Agent for mKGQA
- 7.1 Introduction
- 7.2 The mKGQAgent Architecture
- 7.3 Experimental Setup
- 7.3.1 Large Language Models and Text Embedding Models
- 7.3.2 Implementation of mKGQAgent
- 7.3.3 Baselines
- 7.3.4 Machine Translation of the Input
- 7.4 Evaluation and Analysis
- 7.4.1 English-only Comparison with the Baselines
- 7.4.2 Multilingual Comparison with the Baselines
- 7.4.3 Machine Translation for non-English Questions
- 7.4.4 LLM Calls and Costs
- 7.4.5 Impact of Individual Components on the Quality
- 7.5 Discussion
- 7.6 Conclusion
- 8 Discussion
- 9 Conclusion
- Bibliography
- A A Taxonomy of the Methods to mKGQA