Leveraging LLM GPT-3.5 for Sentiment Analysis: A Comparative Study Across PC and Google Colab
Lavanya B. N, Anitha Rathnam K. V, Abhishek Appaji, K. Kiran, P. Deepa Shenoy, K R Venugopal · 2024
The aim of this study is to inspect sentiment analysis models that have been trained on Twitter corpus by utilising the Large Language Model (LLM) gpt-3-5-turbo-16k version of the Generative Pretrained Transformer (GPT 3.5) model. The Bidirectional long short-term memory (BiLSTM), Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU) and Recurrent Neural Network (RNN) are the trained models, which are used to perform computational tasks on the Google Colab and Personal PC. On the other hand, the training accuracies varied from 51.98% to 99.88% and the test accuracies varied from 50.00% to 75.00%. Every model has its accuracy, recall and f1-score metrics which will show the performance score of that specific model across the platforms. The refined aspects, such as time complexity, memory consumed, and CPU/GPU allocated to find the computational resource expenses on training and testing phases of each model and the issues that were pondered. The study is the first step for the sentiment analysis systems decision makers by showing the best deployment modes via presenting the sentiments classifiers that are used along with different architectures and computing configurations.