Metaheuristics based long short term memory optimization for sentiment analysis

doi:10.1016/j.asoc.2022.109794

2022-12-01

PRE

Mehtab Kiran Suddle

Maryam Bashir *

DOI：10.1016/j.asoc.2022.109794

原文链接

原文求助

摘要

En 中文

A tremendous amount of opinionated data is being added to online platforms every day. Social media has become the primary source of collecting user's opinions about products they have purchased. These online reviews contain important information for the sellers but the information is available in unstructured natural language text. There is a dire need for an automated approach that can extract sentiment from a large number of unstructured reviews. LSTM has shown state-of-the-art results for sentiment analysis in recent years. The performance of LSTM heavily depends on the architectural design. Rather than applying a trial and error approach which can easily mislead the results, an automated optimization technique should be used for architecture related hyperparameter selection. To address this problem, we propose a new framework to optimize the architecture of LSTM using different meta-heuristics. This is the first systematic study of architectural optimization in the context of sentiment analysis using meta heuristics. We have applied Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Firefly, and Cat Swarm Optimization (CSO) for LSTM architecture optimization. The problem of training LSTM has been formulated as an optimization problem and the objective is to maximize F-score. Four benchmark datasets for sentiment analysis have been used in our experiments. PSO and DE show remarkable success in improving the F-score and accuracy. Experimental results demonstrate that the optimal configuration obtained for designing the LSTM architecture using our proposed meta-heuristics significantly improves the accuracy of sentiment analysis.(c) 2022 Elsevier B.V. All rights reserved.

Keyword:

LSTM

Sentiment analysis

Cat swarm optimization

Differential evolution

Firefly

Genetic algorithm

PSO