ALDONA: a hybrid solution for sentence-level aspect-based sentiment analysis using a lexicalised domain ontology and a neural attention model

Abstract

Sentences containing several different polarity aspects cause one of the main problems in sentiment analysis. Depending on an aspect, the same context words can have different effects on its sentiment value. Additionally, the polarity can be influenced by the domain-specific knowledge, showing the necessity to incorporate it into the sentiment classification. In this paper we present a hybrid solution for sentence-level aspect-based sentiment analysis using A Lexicalised Domain Ontology and Neural Attention (ALDONA) model to handle the problems mentioned above. To measure the influence of each word in a given sentence on an aspect s polarity, we introduce the bidirectional context attention mechanism. Moreover, the classification module is designed to handle the sentence s complex structure. Finally, the manually created lexicalised domain ontology (represented in OWL) is integrated to exploit the field-specific knowledge. Computational results obtained on a benchmark data set based on Web reviews have shown ALDONA s ability to outperform several state-of-the-art models and stress its contribution to aspect-based sentiment classification.