Powered by OpenAIRE graph
Found an issue? Give us feedback
addClaim

This Research product is the result of merged Research products in OpenAIRE.

You have already added 0 works in your ORCID record related to the merged Research product.

Exploring transformers for aspect based sentiment analysis

Authors: Poudel, Roshan;

Exploring transformers for aspect based sentiment analysis

Abstract

Nos últimos anos, a análise de sentimentos ganhou atenção significativa por sua ampla gama de aplicações em vários domínios. A análise de sentimentos baseada em aspetos (ABSA) é uma tarefa desafiadora dentro da análise de sentimentos que visa identificar a polaridade do sentimento em relação a aspetos ou atributos específicos de uma entidade alvo em um determinado texto. Os transformadores, um tipo de arquitetura de rede neural profunda, mostraram resultados promissores em muitas tarefas de processamento de linguagem natural (NLP), incluindo análise de sentimento. Esta dissertação explora a eficácia de um modelo BERT+BiLSTM+CRF para ABSA e investiga o impacto dos tamanhos dos modelos e congelamento de camadas. Foram realizadas várias experiências usando diferentes conjuntos de dados ABSA, comparando os resultados com modelos de última geração existentes. Os resultados indicam que aumentar o tamanho do modelo não é necessariamente a melhor abordagem para melhorar o desempenho, e congelar um subconjunto de camadas pode levar a resultados comparáveis com requisitos computacionais reduzidos. O estudo também destaca o impacto dos métodos de pré-treino e conjuntos de dados em tarefas posteriores. O sistema end-to-end desenvolvido é modular, permitindo a substituição do modelo BERT por qualquer outro modelo baseado em transformador baseado no caso de uso. A pesquisa contribui para a compreensão de modelos baseados em transformadores para ABSA e fornece indicadores para estudos futuros neste campo.

In recent years, sentiment analysis has gained significant attention for its wide range of applications in various domains. Aspect-based sentiment analysis (ABSA) is a challenging task within sentiment analysis that aims to identify the sentiment polarity towards specific aspects or attributes of a target entity in a given text. Transformers, a type of deep neural network architecture, have shown promising results in many natural language processing (NLP) tasks, including sentiment analysis. This dissertation explores the effectiveness of a BERT+BiLSTM+CRF model for ABSA and investigates the impact of model sizes and layer freezing. Several experiments are performed using different ABSA datasets, comparing the results with existing state-of-the-art models. The findings indicate that increasing model size is not necessarily the best approach to improve performance, and freezing a subset of layers can lead to comparable results with reduced computational requirements. The study also highlights the impact of pretraining methods and datasets in downstream tasks. The developed end-to-end system is modular, allowing for the replacement of BERT with any other transformer-based model based on the use case. The research contributes to the understanding of transformer-based models for ABSA and provides insights for future studies in this field.

Mestrado em Ciência de Dados

Related Organizations
Keywords

Sentiment analysis, Aspect-based sentiment analysis (ABSA), Transformers, Deep learning, Natural language processing (NLP)

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Upload OA version
Are you the author of this publication? Upload your Open Access version to Zenodo!
It’s fast and easy, just two clicks!