Serialized Multi-Layer Multi-Head Feature Location For Sentiment Analysis In Customer Reviews

Abstract

Sentiment analysis seeks to identify the sentiment orientation of a text item (sentence or document), however finer-grained sentiment categorization is the best option because many real-world applications demand a deeper level of analysis. The process of determining emotional polarity for component phrases in a sentence is known as Aspect-level Sentiment Classification (ALSC). In the ALSC job, the high dimensionality problem is typically encountered, and methods for selecting features are presented as a means of addressing this issue. This work proposes a feature extraction strategy based on Serialized Multi-layer Multi-Head Attention (SMMHA) and Bidirectional Encoder Representations from Transformers (BERT) for classification. For feature selection, the Chaotic Cuckoo Search Optimization (CCSO) algorithm is presented. The assessment of every feature is the first step of the CCSO algorithm search, which then chooses the feature with the best efficiency. Semantics, sentiment, readability, structure, and grammar are all included. The CCSO method is a heuristic search algorithm that draws inspiration from a synopsis of the cuckoo's parasitic and reproductive habits. The nests with greater accuracy values are chosen as the number of iterations rises and employed in the sentiment prediction for aspect-based analysis. In addition to producing interactive semantic information between the aspect word and the context, the feature extraction approach captures the context's long-term reliance. For testing reasons, the Amazon Customer Review Dataset is obtained straight from Amazon. Sentiment analysis accuracy, recall, F1-score, and precision are used to gauge the results.