@inproceedings{10.1007/978-3-031-28244-7_43,
author = {Mamta and Ekbal, Asif},
title = {Service Is Good, Very Good Or Excellent? Towards Aspect Based Sentiment Intensity Analysis},
year = {2023},
isbn = {978-3-031-28243-0},
publisher = {Springer-Verlag},
address = {Berlin, Heidelberg},
url = {https://doi.org/10.1007/978-3-031-28244-7_43},
doi = {10.1007/978-3-031-28244-7_43},
abstract = {Aspect-based sentiment analysis (ABSA) is a fast-growing research area in natural language processing (NLP) that provides more fine-grained information, considering the aspect as the fundamental item. The ABSA primarily measures sentiment towards a given aspect, but does not quantify the intensity of that sentiment. For example, intensity of positive sentiment expressed for service in service is good is comparatively weaker than in service is excellent. Thus, aspect sentiment intensity will assist the stakeholders in mining user preferences more precisely. Our current work introduces a novel task called aspect based sentiment intensity analysis (ABSIA) that facilitates research in this direction. An annotated review corpus for ABSIA is introduced by labelling the benchmark SemEval ABSA restaurant dataset with the seven (7) classes in a semi-supervised way. To demonstrate the effective usage of corpus, we cast ABSIA as a natural language generation task, where a natural sentence is generated to represent the output in order to utilize the pre-trained language models effectively. Further, we propose an effective technique for the joint learning where ABSA is used as a secondary task to assist the primary task, i.e. ABSIA. An improvement of 2 points is observed over the single task intensity model. To explain the actual decision process of the proposed framework, model explainability technique is employed that extracts the important opinion terms responsible for generation (Source code and the dataset has been made available on , )},
booktitle = {Advances in Information Retrieval: 45th European Conference on Information Retrieval, ECIR 2023, Dublin, Ireland, April 2–6, 2023, Proceedings, Part I},
pages = {685–700},
numpages = {16},
keywords = {Joint learning, Explainability, Aspect sentiment intensity, Sentiment analysis, Absa, Generation},
location = {Dublin, Ireland}
}