Ontology Enrichment for Aspect-Oriented Sentiment Analysis with Deep Learning Using Logical Concept Analysis

Abstract

Aspect-oriented sentiment analysis is a problem that have been attracting much attention from the research community. In the past, using ontologies to represent domain knowledge helped increase the performance of this task. Recently, with the rapid development of deep learning techniques, combination of deep learning with ontology has become very promising. A natural approach is to use aspects represented by ontologies to form input vectors s for deep learning models. However, ontologies often only represent concrete concepts, whereas machine learning techniques often focus on latent features. To address this, we propose using Logical Concept Analysis (LCA), an extension of Formal Concept Analysis (FCA), to enrich the ontology by generating more abstract concepts from existing concrete concepts. We have applied this approach to real data and achieved promising results.