Does Coreference Resolution Improve Aspect Based Sentiment Analysis?

Aspect-Based Sentiment Analysis (ABSA) has generally focused on extracting explicit opinion targets and classifying them into polarities and categories. Most approaches ignore implicitly expressed opinions, even though they make up a significant part of language; in fact, approximately 25% of the ta...

Descripción completa

Detalles Bibliográficos
Autor: Ryhänen, Rosa-Maria Kristiina
Tipo de recurso: tesis de maestría
Fecha de publicación:2022
País:España
Institución:Universidad del País Vasco
Repositorio:Addi. Archivo Digital para la Docencia y la Investigación
OAI Identifier:oai:addi.ehu.eus:10810/61818
Acceso en línea:http://hdl.handle.net/10810/61818
Access Level:acceso abierto
Palabra clave:aspect-based sentiment analysis
coreference resolution
opinion target extraction
aspect category detection
Descripción
Sumario:Aspect-Based Sentiment Analysis (ABSA) has generally focused on extracting explicit opinion targets and classifying them into polarities and categories. Most approaches ignore implicitly expressed opinions, even though they make up a significant part of language; in fact, approximately 25% of the targets in the SemEval ABSA 2016 English restaurant reviews (Pontiki et al., 2016) are implicit and are not taken into consideration when training a model. We propose to solve a part of the implicit targets with coreference resolution in order to improve two ABSA tasks: opinion target extraction and aspect category detection. Our results suggest that coreference resolution helps to perform opinion target extraction and aspect category detection, when the latter is handled as a multi-label classification task. The data and code are publicly available on GitHub https://github.com/rosamariaryh/absa-coref