Open Access

ARTICLE

Opinion Mining on Movie Reviews Based on Deep Learning Models

Mian Muhammad Danyal¹, Muhammad Haseeb¹, Sarwar Shah Khan^2,*, Bilal Khan¹, Subhan Ullah¹

1 Department of Computer Science, City University of Science & Technology, Peshawar, 25000, Pakistan
2 Department of Computer Science, Iqra National University, Swat, 19200, Pakistan

* Corresponding Author: Sarwar Shah Khan. Email:

Journal on Artificial Intelligence 2024, 6, 23-42. https://doi.org/10.32604/jai.2023.045617

Received 02 September 2023; Accepted 15 December 2023; Issue published 31 January 2024

Abstract

Movies reviews provide valuable insights that can help people decide which movies are worth watching and avoid wasting their time on movies they will not enjoy. Movie reviews may contain spoilers or reveal significant plot details, which can reduce the enjoyment of the movie for those who have not watched it yet. Additionally, the abundance of reviews may make it difficult for people to read them all at once, classifying all of the movie reviews will help in making this decision without wasting time reading them all. Opinion mining, also called sentiment analysis, is the process of identifying and extracting subjective information from textual data. This study introduces a sentiment analysis approach using advanced deep learning models: Extra-Long Neural Network (XLNet), Long Short-Term Memory (LSTM), and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM). XLNet understands the context of a word from both sides, which is helpful for capturing complex language patterns. LSTM performs better in modeling long-term dependencies, while CNN-LSTM combines local and global context for robust feature extraction. Deep learning models take advantage of their ability to extract complex linguistic patterns and contextual information from raw text data. We carefully cleaned the IMDb movie reviews dataset with the goal of optimizing the results of models used in the experiment. This involves eliminating unnecessary punctuation, links, hashtags, stop words, and duplicate reviews. Lemmatization is also used for keeping consistent word forms. This cleaned IMDb dataset is evaluated on the proposed model for sentiment analysis in which XLNet performs well achieving an impressive 93.74% accuracy on the IMDb Dataset. The findings highlight the effectiveness of deep learning models in improving sentiment analysis, showing its potential for wider applications in natural language processing.

---

Keywords

---