E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Search command : Author="Στεφανίδης" And Author="Κωνσταντίνος"

Current Record: 22 of 824

[Add to Basket]

Identifier

000460702

Title

Question answering over CIDOC-CRM based knowledge graphs

Alternative Title

Απάντηση ερωτήσεων επί γνωσιακών γράφων που βασίζονται στο CIDOC -CRM

Author

Γουνάκης, Νικόλαος Ν.

Thesis advisor

Τζίτζικας, Γιάννης

Reviewer

Πλεξουσάκης, Δημήτριος
Μαγκούτης, Κωνσταντίνος

Abstract

CIDOC-CRM is a standard for documenting cultural heritage based on events that enables semantic interoperability among different sources of data in the Cultural Heritage (CH) domain. Despite the existence of several Knowledge Graphs (KGs) that use CIDOC-CRM, the problem of Question Answering (QA) over such graphs has not been explored extensively. Therefore, in this thesis we propose and evaluate a Radius-based QA pipeline over CIDOC-CRM KGs for mostly answering single-entity factoid questions ,while also covering confirmation questions. Specifically, we present a generic QA pipeline that consists of various models and methods, such as a keyword search model for identifying the entity of the question (and linking it to the KG), methods that rely on path expansion for building subgraphs of different radius (or depths) starting from the identified entity, i.e., for providing a context, and pre-trained neural models (based on BERT) for answering the question using the given context. Furthermore, since there are no available benchmarks over CIDOC-CRM KGs, we create (by using a real KG) an evaluation benchmark with 10,000 questions, i.e., 5,000 single-entity factoid, 2,500 comparative and 2,500 confirmation questions. For evaluating the QA pipeline, we use the 5,000 single-entity factoid questions and the 2,500 confirmation. Finally, we create a simple web application that enables the QA task to the users by utilizing the pipeline. Regarding the evaluation results, the QA pipeline achieves satisfactory results for factoid questions both in the entity recognition step (78% accuracy) and in the QA process (51% F1 score), while in confirmation 54% accuracy for entity detection and 76% accuracy for biased methods in the QA process, indicating the need for radius prediction.

Language

English

Subject

Natural language processing

Γνωσιακοί γράφοι

Επεξεργασία φυσικής γλώσσας

Issue date

2023-12-01

Collection

School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses

Type of Work--Post-graduate theses

Permanent Link