Quantum-Classical Sentiment Analysis
Mario Bifulco, Luca Roversi
TL;DR
The paper tackles the training-time bottleneck in expressive NLP models by evaluating a hybrid quantum-classical classifier on adiabatic quantum computing and comparing it to a classical CPLEX solver and a Transformer baseline using $QUBO$-formulations. It presents a Quantum Support Vector Machine approach with TweetEval and Sentence-BERT embeddings, and introduces a novel QSplit decomposition to maximize quantum processing unit ($QPU$) usage. Key findings show that while Transformer models deliver higher accuracy (e.g., RoBERTa ~94.3%), the quantum approach achieves competitive F1 scores (~76%), and training times on the quantum side can be significantly faster (~39.2s) than classical solvers (~101.9s). The study demonstrates a practical pathway to leverage quantum acceleration for NLP optimization in resource-constrained settings, while highlighting embedding challenges and the need for refined partitioning strategies to balance speed and solution quality.
Abstract
In this study, we initially investigate the application of a hybrid classical-quantum classifier (HCQC) for sentiment analysis, comparing its performance against the classical CPLEX classifier and the Transformer architecture. Our findings indicate that while the HCQC underperforms relative to the Transformer in terms of classification accuracy, but it requires significantly less time to converge to a reasonably good approximate solution. This experiment also reveals a critical bottleneck in the HCQC, whose architecture is partially undisclosed by the D-Wave property. To address this limitation, we propose a novel algorithm based on the algebraic decomposition of QUBO models, which enhances the time the quantum processing unit can allocate to problem-solving tasks.