Combining Query Rewriting and Knowledge Graph Embeddings for Complex Query Answering
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3072148Utgivelsesdato
2023-06-01Metadata
Vis full innførselSamlinger
- Master theses [197]
Sammendrag
The field of complex query answering using Knowledge Graphs (KGs) has seen substantial advancements in recent years, primarily through the utilization of Knowledge Graph Embeddings (KGEs). However, these methodologies often stumble when faced with intricate query structures that involve multiple entities and relationships. This thesis primarily investigates the potential of integrating query rewriting techniques into the KGE query answering process to improve performance in such situations. Guided by a TBox, a schema that describes the concepts and relationships in the data from Description Logics, query rewriting translates a query into a union of rewritten queries that can potentially widen the prediction scope for KGEs. The thesis uses the PerfectRef algorithm for facilitating query rewriting, aiming to maximize the scope of query response and enhance prediction capabilities. Two distinct datasets were employed in the study: The Family Dataset, a subset of Wikidata, and DBPedia15k, a subset of DBPedia. The effectiveness of the proposed methodology was evaluated against these datasets using different KGE models, in our case TransE, DistMult, BoxE, RotatE, and CompGCN. The results demonstrate a notable improvement in complex query answering when query rewriting is used for both The Family dataset and DBPedia15k. Furthermore, the amalgamation of query rewriting and KGE predictions yielded a performance boost for The Family dataset. However, the same was not observed for DBPedia15k, likely due to discrepancies and errors present within DBPedia15k compared to the Full DBPedia KG used for validation in our framework. This research suggests that query rewriting, as a pre-processing step for KGE prediction, can enhance the performance of complex query answering, mainly when the dataset is not fully entailed. This study provides important insights into the potential and limitations of integrating query rewriting with KGEs. It may serve as a guidepost for future research to improve the complex query answering when a TBox is available.
Utgiver
The University of BergenOpphavsrett
Copyright the Author. All rights reservedBeslektede innførsler
Viser innførsler beslektet ved tittel, forfatter og emneord.
-
FILT - Filtering Indexed Lucene Triples
Stuhr, Magnus (Master thesis, 2012-05-31)The Resource Description Framework (RDF) is the W3C recommended standard for data on the semantic web, while the SPARQL Protocol and RDF Query Language (SPARQL) is the query language that retrieves RDF triples by subject, ... -
Drawing visual query images : use, users and usability of query by drawing interfaces for content based image retrieval systems
Hove, Lars-Jacob (Doctoral thesis, 2010-12-14)Query by Drawing (QBD) is an approach to Content Based Image Retrieval (CBIR) systems where users express their image needs by drawing an image representative of the images they wish to retrieve. CBIR is based on comparison ... -
Characterizing Triacylglycerol in Cell Culture: Can LCMS replace HPTLC when assessing the effect of metformin in salmon in-vitro fatty liver model?
Pasha, Monirul Hoque (Master thesis, 2017-10-25)Fatty liver, a health complication for aquacultured fish is developed by the deposition of triacylglycerol (TAG) within liver cells. Metformin has been proven to be an effective drug for ameliorating fatty liver in human ...