Object-Level Sentiment Analysis Use a Language Model

Abstract

Sentiment analysis remains a prominent area of research in the natural language processing (NLP) community and holds significant practical value in domains such as commerce and education. Most existing approaches evaluate sentiments for a single object or product, typically categorizing them as positive or negative. However, when a text involves comparisons between multiple objects, it can be challenging to identify which sentiment or emotion is associated with which object. Few studies have addressed this issue, often stopping at evaluating emotions at the sentence level or for individual words related to aspects or objects. This study proposes an object-level sentiment analysis problem that produces a set of pairs or triples consisting of an object, aspect, and sentiment. Additionally, in texts expressing opinions or comments on a specific aspect, the aspect may be implied through references to the object without being explicitly mentioned. Identifying such implicit aspects is crucial, as it ensures no loss of information and enhances the efficiency of extraction of information in object-level sentiment analysis. The integration of implicit aspect identification and object-level sentiment analysis is the primary focus of this research. In recent years, many language models have been developed and effectively applied to various NLP tasks. Therefore, to address the proposed challenges, this study utilizes deep learning that incorporates language models combined with NLP methods such as parsing and dependency analysis, to achieve the desired output. Using language model and NLP techniques automatically generate training data for the learning model. The proposed method achieves an accuracy of 90%, making a substantial contribution to the field of NLP.