Open Access

ARTICLE

Aspect Extraction Approach for Sentiment Analysis Using Keywords

Nafees Ayub¹, Muhammad Ramzan Talib^1,*, Muhammad Kashif Hanif¹, Muhammad Awais²

1 Department of Computer Science, Government College University, Faisalabad, 38000, Pakistan
2 Department of Software Engineering, Government College University, Faisalabad, 38000, Pakistan

* Corresponding Author: Muhammad Ramzan Talib. Email:

Computers, Materials & Continua 2023, 74(3), 6879-6892. https://doi.org/10.32604/cmc.2023.034214

Received 09 July 2022; Accepted 22 September 2022; Issue published 28 December 2022

Abstract

Sentiment Analysis deals with consumer reviews available on blogs, discussion forums, E-commerce websites, and App Store. These online reviews about products are also becoming essential for consumers and companies as well. Consumers rely on these reviews to make their decisions about products and companies are also very interested in these reviews to judge their products and services. These reviews are also a very precious source of information for requirement engineers. But companies and consumers are not very satisfied with the overall sentiment; they like fine-grained knowledge about consumer reviews. Owing to this, many researchers have developed approaches for aspect-based sentiment analysis. Most existing approaches concentrate on explicit aspects to analyze the sentiment, and only a few studies rely on capturing implicit aspects. This paper proposes a Keywords-Based Aspect Extraction method, which captures both explicit and implicit aspects. It also captures opinion words and classifies the sentiment about each aspect. We applied semantic similarity-based WordNet and SentiWordNet lexicon to improve aspect extraction. We used different collections of customer reviews for experiment purposes, consisting of eight datasets over seven domains. We compared our approach with other state-of-the-art approaches, including Rule Selection using Greedy Algorithm (RSG), Conditional Random Fields (CRF), Rule-based Extraction (RubE), and Double Propagation (DP). Our results have shown better performance than all of these approaches.

---

Keywords

---