Research map / Label-efficient and robust learning / Recognition and long-tail
QPEN: Quantum Projection and Quantum Entanglement Enhanced Network for Cross-Lingual Aspect-Based Sentiment Analysis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes QPEN, a quantum-inspired network that projects aspects into a complex Hilbert space and shares language-specific features via a simulated quantum entanglement module on top of mBERT/XLM-R for cross-lingual ABSA.
Problem
Cross-lingual aspect-based sentiment analysis, where a model trained on English must transfer to other languages (FR, SP, DU, RU) without relying on translation quality or label projection, and low-resource languages are under-represented in multilingual pre-training.
Approach
Formulates each aspect as a quantum superposition over positive, neutral, and negative sentiment basis states in a complex 3-dimensional Hilbert space (quantum projection), and uses a five-qubit entangled state built with Hadamard and CNOT gates so that language-specific features are shared across languages without transmission (quantum entanglement module), with both representations concatenated
Key results
- QPEN with mBERT improves average F1 over the prior state of the art (ACS-DISTILL-M) by 1.03 points across four target languages (FR, SP, DU, RU).
- QPEN with XLM-R improves average F1 by 2.31 points over ACS-DISTILL-M, reaching 65.79 average.
- Ablation: quantum projection and quantum entanglement each improve over the ACS baseline on both mBERT and XLM-R.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use QPEN instead of ACS-DISTILL-M | mBERT · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 55.79 -> 56.97 | Helped |
| use QPEN instead of the strongest baseline | XLM-R · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 63.45 -> 65.79 | Helped |
| add quantum projection to ACS (ablation) | mBERT · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 53.23 -> 54.58 | Helped |
| add quantum entanglement to ACS (ablation) | mBERT · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 53.23 -> 54.83 | Helped |
| add quantum projection to ACS (ablation) | XLM-R · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 62.59 -> 64.33 | Helped |
| add quantum entanglement to ACS (ablation) | XLM-R · SemEval-2016 average (FR, SP, DU, RU) | F1-score: 62.59 -> 63.72 | Helped |
Surprising findings
- Quantum projection benefits XLM-R more than mBERT, while quantum entanglement benefits mBERT more than XLM-R.
Limitations
- Experiments are simulated on a classical computer; the quantum circuit is not run on quantum hardware.
- Evaluated only on SemEval-2016 with four target languages (Turkish excluded), and the entanglement module is designed for exactly five languages.
- Fails on some cases involving price comparisons and difficult sentiment identification, per the error study.
Experimental setup
Datasets: SemEval-2016 ABSA
Benchmarks: SemEval-2016 ABSA
Metrics: F1
Baselines: ZERO-SHOT TRANSLATION-TA BILINGUAL-TA TRANSLATION-AF BILINGUAL-AF ACS ACS-DISTILL-S ACS-DISTILL-M
Builds on: ACS mBERT XLM-R
Closest papers by meaning
- Quantum-Inspired Neural Network with Runge-Kutta Method (AAAI 2024 · similarity 0.84)
- ATP: Adaptive Threshold Pruning for Efficient Data Encoding in Quantum Neural Networks (CVPR 2025 · similarity 0.82)
- Quantum Interference Model for Semantic Biases of Glosses in Word Sense Disambiguation (AAAI 2024 · similarity 0.82)
- QuanTA: Efficient High-Rank Fine-Tuning of LLMs with Quantum-Informed Tensor Adaptation (NeurIPS 2024 · similarity 0.81)
- Exploring channel distinguishability in local neighborhoods of the model space in quantum neural networks (ICLR 2025 · similarity 0.81)
- Quantum-inspired Non-homologous Representation Constraint Mechanism for Long-tail Senses of Word Sense Disambiguation (AAAI 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.