APPLICATION OF HYBRID QUANTUM-CLASSICAL MACHINE LEARNING METHODS IN SENTIMENT ANALYSIS

• Main Author: Rifat Abiwardani, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74188
• Institution: Institut Teknologi Bandung

The development of quantum computing has prompted researchers to solve natural language processing tasks using quantum algorithms. As high-level libraries for implementing DisCoCat Quantum Natural Language Processing (QNLP) using the Variational Quantum Algorithms (VQA) framework for hybrid quantum-classical quantum machine learning have been developed, multiple research has been con- ducted in this field. One such paper is Ruskanda et al., 2022 where sentiment anal- ysis is performed using QNLP framework. This Final Project expands on that work by exploring alternative ansatze for VQA. Namely, this work explores SimpleSA, ? Sim15, and GeneralQC ansatze and compares them to Instantaneous Quantum Poly- ? time (IQP) ansatz. Hyperparameter experimentation is also done to determine the best settings of noun parameterization, number of layers, and maximum number of Simultaneous Perturbation Stochastic Approximation (SPSA) iterations. The per- formance of each ansatz experiment is measured based on accuracy, fit time, and coherence. This work reformulates a complete QNLP pipeline for performing sen- timent analysis based on the QNLP pipeline proposed in Kartsaklis et al., 2021 and the VQA framework. It was found that SimpleSA, Sim15, and GeneralQC outper- form IQP ansatz. Furthermore, results of this work show that GeneralQC ansatz performs best and is the most robust at 90.00% accuracy. Tuning revealed that not parameterizing nouns is highly suitable for sentiment analysis, and that GeneralQC ansatz performs best when only 1 block component layer is used and SPSA is run for 130 iterations.