We present a component-wise ablation study of a quantum natural language processing (QNLP) pipeline for sentiment classification, focusing on how classical and quantum design choices interact under NISQ constraints. Starting from a baseline, we evaluate five different qubit numbers, six sentence embedding strategies, two dimensionality reduction methods, two feature encoding schemes, and three variational ansatzes. Each stage is tested independently while fixing previously optimized components, allowing us to isolate the contribution of each. Experiments are conducted on simulated quantum backends. We find that sentence-level transformer embeddings (MiniLM, MPNet), linear dimensionality reduction (PCA), and simple angle encoding provide substantial accuracy gains without additional training cost. Finally, we introduce a shallow custom ansatz that induces a multiplicative cosine kernel across qubits. While the best overall test accuracy$(87 \%)$is achieved using the RealAmplitude ansatz, the custom ansatz closely follows (86.5 %) while reducing runtime by over$35 \%(277 ~\mathrm{s}$vs. 432 s). This demonstrates a promising trade-off between performance and efficiency, supporting the design of scalable QNLP pipelines under limited quantum resources.
We present a component-wise ablation study of a quantum natural language processing (QNLP) pipeline for sentiment classification, focusing on how classical and quantum design choices interact under NISQ constraints. Starting from a baseline, we evaluate five different qubit numbers, six sentence embedding strategies, two dimensionality reduction methods, two feature encoding schemes, and three variational ansatzes. Each stage is tested independently while fixing previously optimized components, allowing us to isolate the contribution of each. Experiments are conducted on simulated quantum backends. We find that sentence-level transformer embeddings (MiniLM, MPNet), linear dimensionality reduction (PCA), and simple angle encoding provide substantial accuracy gains without additional training cost. Finally, we introduce a shallow custom ansatz that induces a multiplicative cosine kernel across qubits. While the best overall test accuracy(87%)is achieved using the RealAmplitude ansatz, the custom ansatz closely follows (86.5 %) while reducing runtime by over35%(277svs. 432 s). This demonstrates a promising trade-off between performance and efficiency, supporting the design of scalable QNLP pipelines under limited quantum resources.
Abstract Hybrid quantum–classical models for text classification are constrained by the mismatch between high-dimensional sentence embeddings and the limited number of qubits available in Noisy Intermediate-Scale Quantum (NISQ) hardware. Existing approaches rely on fixed dimensionality-reduction techniques such as PCA, which compress embeddings independently of the learning task and often remove semantic information essential for downstream classification. This work addresses this limitation by applying Neural Quantum Embedding (NQE) to sentence-level sentiment analysis. NQE jointly performs nonlinear neural compression and quantum-oriented encoding, producing compact embeddings optimized for variational circuits.We integrate NQE with a Quantum Neural Network based on a three-layer SU(4) ansatz. Experiments on three sentence-level sentiment datasets show that NQE substantially improves hybrid quantum performance, increasing accuracy by up to 38%, with corresponding improvements in precision, recall, and F1-score. Additional experiments with a classical neural network demonstrate that NQE also enhances classical performance, yielding accuracy gains of up to 6.3%. Although classical models achieve higher absolute accuracy due to greater representational capacity, the relative improvement introduced by NQE is significantly larger in the quantum setting, where embedding structure must align with strict hardware constraints.Overall, this study introduces a task-adaptive, learnable embedding mechanism that bridges the representational gap between classical embeddings and quantum circuits. The results show that NQE improves optimization stability, enhances quantum-state separability, and enables more effective hybrid sentiment classification within NISQ limitations
Quantum neural networks (QNNs) have recently attracted remarkable attention as a potential alternative to classical and deep neural network (NN) models particularly under constraints imposed by noisy intermediate-scale quantum (NISQ) devices, which support only shallow circuits with few quantum bits (qubits) and limited variational depth. However, applying QNNs to natural language processing (NLP) remains challenging because text representations are typically high-dimensional, while NISQ classifiers can only encode a small number of features at a time, requiring classical compression to enable stable training and flexible circuit configuration. In addition, many existing studies integrate QNNs into rich-text representation pipelines, where the quantum circuit is often implemented as a variational quantum circuit (VQC). Here, the VQC acts as an additional component within a larger model, and it remains unclear how the quantum component improves the classification task. Therefore, in this work, we propose a fully reproducible and compact QNN pipeline that provides a controlled setting for studying the contribution of the VQC to text classification. The model is trained end-to-end with a lightweight classical compression layer, and the overall study is supported by systematic ablation analyses of the compression layer, the classifier heads under a shared frozen low-dimensional representation, and quantum circuit structure. Our model uses the term frequency-inverse document frequency (TF-IDF) features with linear compression followed by a shallow four-qubit, two-layer VQC, implemented in PyTorch and PennyLane for end-to-end training. It is evaluated on five benchmark text classification datasets against classical baselines, including logistic regression and multilayer perceptron (MLP) models. Experimental results show that the proposed QNN achieves a competitive performance across all evaluated datasets, yielding the best results on three of the five benchmark datasets.
Large language and sentence-embedding models provide rich semantic representations, but their high dimensionality poses a challenge for near-term quantum machine learning (QML), where quantum circuits can process only a limited number of input features. We investigate a hybrid quantum-classical pipeline that transforms high-dimensional sentence embeddings into compact representations for variational quantum classification. The workflow combines a pretrained sentence-embedding model, dimensionality reduction, angle encoding, a variational quantum circuit (VQC), and a classical decision layer. We systematically compare principal component analysis (PCA), neighborhood components analysis (NCA), and linear discriminant analysis (LDA), covering both unsupervised and supervised dimensionality reduction. Using the TREC question-classification dataset, we study the relationship between representation dimensionality, information retention, qubit count, and classification performance. Preliminary PCA experiments reveal a strong information bottleneck: reducing 768-dimensional embeddings to 3, 4, 5, and 8 dimensions retains about 8.2%, 10.2%, 11.9%, and 16.4% of the variance, with corresponding classification accuracies of 50.3%, 51.2%, 57.9%, and 63.4%. In contrast, supervised reduction is substantially more efficient. LDA reaches 85.3% accuracy and NCA reaches 83.1% using only 5 dimensions, under a leakage-free cross-validation protocol, compared with 85.1% for a full 384-dimensional classical baseline. These results indicate that supervised dimensionality reduction can preserve task-relevant information far more effectively than variance-based compression, making compact representations a promising route toward practical hybrid quantum-classical NLP models.
Combining variational quantum circuits with classical neural network models is a potential direction for quantum natural language processing (QNLP) in the NISQ (Noisy Intermediate-Scale Quantum) era. We propose a hybrid classical-quantum neural network that enhances a transformer architecture in two key ways: first, by incorporating optimized variational quantum circuits into the self-attention and multilayer perceptron components, and second, by integrating a hybrid convolutional module for effective dimensionality reduction which can reduce the quantum resources required for subsequent layers and enhances the model’s trainability by mitigating issues like the barren plateau phenomenon, thus making it feasible to construct a deeper and more powerful network architecture. Our experiments on three widely-used sentiment analysis datasets demonstrate that the proposed model achieves comparable or even superior performance in terms of accuracy and F1 score compared to classical transformer-based sentiment classification models with similar structures, while reducing the number of parameters through the utilization of quantum advantages.
We study a \emph{QDisCoCirc}-inspired, chunked diagram-to-circuit quantum natural language processing (QNLP) model for three-class sentiment classification of financial texts. In our classical simulations, we keep the Hilbert-space dimension manageable by decomposing each sentence into short contiguous chunks. Each chunk is mapped to a shallow quantum circuit, and the resulting Bloch vectors are used as a sequence of quantum tokens. Simple averaging of chunk vectors ignores word order and syntactic roles. We therefore add a small Transformer encoder over the raw Bloch-vector sequence and attach a CCG-based type embedding to each chunk. This hybrid design preserves physically interpretable semantic axes of quantum tokens while allowing the classical side to model word order and long-range dependencies. The sequence model improves test macro-F1 over the averaging baseline and chunk-level attribution further shows that evidential mass concentrates on a small number of chunks, that type embeddings are used more reliably for correctly predicted sentences. For real-world quantum language processing applications in finance, future key challenges include circuit designs that avoid chunking and the design of inter-chunk fusion layers.
This paper investigates sentiment classification using QNLP within the Variational Quantum Algorithms (VQA) framework, which offers a practical approach for current hardware limitations by optimizing parameterized quantum circuits through classical computation. Designing an effective VQA ansatz is a key challenge because it directly influences the model’s expressiveness and trainability. To address this, a new ansatz called the GeneralQC ansatz is proposed by combining the preliminary SimpleSA ansatz with the circuit-centric quantum classifier, aiming to enhance performance by integrating the strengths of both methods. Three instantiations of the GeneralQC ansatz—H-CRx, Rx-CNOT, and Rx-CRz—are explored under the SimpleSA strategy of defining noun word boxes as fixed identity gates and are compared with both the baseline SimpleSA ansatz and versions with noun parameterization. The findings reveal that the Rx-CNOT and Rx-CRz models with non-parameterized nouns significantly outperform the SimpleSA ansatz, achieving an accuracy of 96.91%.