A Self-matching Training Method with Annotation Embedding Models for Ontology Subsumption Prediction

TL;DR

The self-matching training method with InME outperforms the existing ontology embeddings for the GO and FoodOn ontologies and the method with the concatenation of CoME and OWL2Vec* outperforms them for the HeLiS ontology.

Abstract

Recently, ontology embeddings representing entities in a low-dimensional space have been proposed for ontology completion. However, the ontology embeddings for concept subsumption prediction do not address the difficulties of similar and isolated entities and fail to extract the global information of annotation axioms from an ontology. In this paper, we propose a self-matching training method for the two ontology embedding models: Inverted-index Matrix Embedding (InME) and Co-occurrence Matrix Embedding (CoME). The two embeddings capture the global and local information in annotation axioms by means of the occurring locations of each word in a set of axioms and the co-occurrences of words in each axiom. The self-matching training method increases the robustness of the concept subsumption prediction when predicted superclasses are similar to subclasses and are isolated to other entities in an ontology. Our evaluation experiments show that the self-matching training method with InME outperforms the existing ontology embeddings for the GO and FoodOn ontologies and that the method with the concatenation of CoME and OWL2Vec* outperforms them for the HeLiS ontology.

Figures (6)

• Figure 1: The MRRs of Word2Vec for logical structures and annotations on GO.
• Figure 2: The framework of a self-matching training method with InME and CoME. (a) The inverted-index and co-occurrence matrices, $X^{\rm{global}}$ and $X^{\rm{local}}$, are constructed from annotation axioms. (b) The word embeddings of $H_{i}^{\ast}$ are obtained from the matrices compressed with an autoencoder and entity embeddings $V_{\rm{word}}^{\ast}$ are transformed by averaging the word embeddings. (c) An RF classifier is trained by the self-matching training method with InME or CoME.
• Figure 3: The flowchart of the conventional RF training and self-matching training.
• Figure 4: The MRRs of InME, CoME, and InME $||$ CoME.
• Figure 5: The t-SNE visualizations of the embeddings OWL2Vec*, InME, and CoME on GO, FoodOn, and HeLiS.