dice-group / tab2onto

Tab2Onto: Unsupervised Semantification with Knowledge Graph Embeddings

Tab2Onto: Unsupervised Semantification with Knowledge Graph Embedding

Overview:

We propose, Tab2Onto, an unsupervised approach for learning ontologies from tabular data. Our approach includes five steps as shown in Figure 1:

(a) Data preprocessing: Given input tabular data (CSV), we first preprocess the data and transform it into a knowledge graph (RDF triples). Each triple describes information about an entity in form <subject, predicate, object>

(b) Knowledge Graph Embedding: in this step, we represent entitiens and their relations into one semantic space.

(c) Clustering: we use a density-based clustering approach to detect clusters of entites. Each cluster contains entities with similar properties. (d) Human-In-The-Loop: we incorporate a domain expert as a human-in-the-loop to label clusters based on the properities of its entities.

(e) Finally, we populate the assigned label (i.e, class) to all entities within the same cluster.

Tab2Onto pipleine for our semantification process

---

Installation:

please install them from requirements.txt file via pip install -r requirements.txt

---

Dataset:

• FB15k-237: this dataset is a subset of Freebase Knowledge Graph contains $310,116$ triples with $14,951$ entities and $237$ relations. The source can be found in data folder. we evaluated our approach on FB15K dataset to assess the performance for predicting types of entities (e.g. movie, person , organization) using embedding-based clustering. As an example of {transE embeddings, hdbscan clustering} results using t-SNE projection. We plotted entities in six types (education, film, location, music, people, and soccer). It's clearly seen that, entities with same type (e.g. film --in orange color--), cluster well based on their embeddings representation.

t-SNE visualization of semantification process on FB15k-237 with TransE embedding.

For more visualization results with different embeddings, please check src/Figures

• Lymphography Data: we investigated our full pipeline on the SML-Bench dataset, Lymphography. We processed all five steps to convert orginal lymphography data from tabular format to an ontology. The learned ontology are saved as an OWL in RDF/XML format.

---

How to run:

There are two folders for our approach implementation:

• src: contains the source-code in Python 3.6
• notebooks: contains three notebooks for our experiment on FB15k-237 and Lymphography datasets.

---

Aknowledgment:

This work has been supported by the German Federal Ministry for Economic Affairs and Climate Action (BMWK) within the project RAKI under the grant no 01MD19012B and by the German Federal Ministry of Education and Research (BMBF) within the project DAIKIRI under the grant no 01IS19085B.

---

Citation

@INPROCEEDINGS
{zahera2022tab2onto, 
author = "Hamada M. Zahera, Stefan Heindorf, Stefan Balke, Jonas Haupt, 
Martin Voigt, Carolin Walter, Fabian Witter and Axel-Cyrille Ngonga Ngomo", 
title = "Tab2Onto: Unsupervised Semantification with Knowledge Graph Embeddings",
booktitle = "The Semantic Web: ESWC 2022 Satellite Events", 
year = "2022", series = "Springer"}

If you have any further questions/suggestions, please contact [email protected]