Title: A comprehensive and comparative analysis of deep learning models for textual sentiment analysis

Authors: Leyla R. Mammadova

Addresses: Institute of Information Technology of Azerbaijan National Academy of Sciences, 9, B. Vahabzade str., Baku, AZ1141, Azerbaijan Republic

Abstract: Analysing public opinion may provide us with important insights. Sentiment analysis is a textual data analysis technique that identifies subjective information expressed by people or groups, including views and emotions. By advancing natural language processing (NLP) and deep learning approaches, sentiment analysis advances our comprehension of human language. This study provides a thorough evaluation and comparative analysis of various deep learning models, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and gated recurrent units (GRUs), and their bidirectional variants. An analysis with three datasets that are accessible to the public is achieved: The imdb_reviews, Twitter Sentiment Dataset, and Emotions dataset. The accuracy performance of six well-known deep learning models is assessed. Experimental results demonstrate that bidirectional architectures perform generally better than their unidirectional equivalents. The bidirectional models consistently achieved the highest accuracy across different datasets.

Keywords: sentiment analysis; RNN; recurrent neural network; LSTM; GRU; gated recurrent unit; bidirectional RNN; bidirectional LSTM; bidirectional GRU.

DOI: 10.1504/IJDATS.2025.150911

International Journal of Data Analysis Techniques and Strategies, 2025 Vol.17 No.4, pp.328 - 344

Received: 28 Jun 2024
Accepted: 13 Aug 2024
Published online: 05 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article