Quantum-inspired Interpretable Deep Learning Architecture for Text Sentiment Analysis

TL;DR

This paper addresses the challenge of integrating diverse semantic information with interpretability in text sentiment analysis. It introduces a quantum-inspired architecture (QITSA) that represents text via density matrices and complex word embeddings, leveraging a quantum embedding layer, LSTM-attention feature extraction, Q-Attention fusion, and 2D-CNNs for final classification. The approach combines semantic and sentiment signals through amplitudes and phases, achieving competitive or superior accuracy across five binary sentiment datasets, with notable interpretability stemming from the QM-inspired representation. The work demonstrates that quantum-inspired representations and attention-guided fusion can improve performance while providing mechanisms for visualization and interpretation, suggesting practical impact for robust, explainable NLP models.

Abstract

Text has become the predominant form of communication on social media, embedding a wealth of emotional nuances. Consequently, the extraction of emotional information from text is of paramount importance. Despite previous research making some progress, existing text sentiment analysis models still face challenges in integrating diverse semantic information and lack interpretability. To address these issues, we propose a quantum-inspired deep learning architecture that combines fundamental principles of quantum mechanics (QM principles) with deep learning models for text sentiment analysis. Specifically, we analyze the commonalities between text representation and QM principles to design a quantum-inspired text representation method and further develop a quantum-inspired text embedding layer. Additionally, we design a feature extraction layer based on long short-term memory (LSTM) networks and self-attention mechanisms (SAMs). Finally, we calculate the text density matrix using the quantum complex numbers principle and apply 2D-convolution neural networks (CNNs) for feature condensation and dimensionality reduction. Through a series of visualization, comparative, and ablation experiments, we demonstrate that our model not only shows significant advantages in accuracy and efficiency compared to previous related models but also achieves a certain level of interpretability by integrating QM principles. Our code is available at QISA.

Figures (13)

• Figure 1: Illustration of Quantum-Inspired Language Representation. (a) Synonymy corresponding to quantum superposition principle. (b) Embedding semantic information into complex numbers (ignoring multimodal information). (c) Left: Embedding both semantic and sentiment information. Right: Bloch Sphere Representation for the QITSA.
• Figure 2: The QM principle word embedding layer. (a) The principle of quantum complex numbers integrating multiple semantic information, and (b) Euler's formula for calculating the imaginary and real parts of the semantic vector matrix.
• Figure 3: Overall process framework and detailed structure diagram. (a)(b)(c) jointly form the common framework of the article.
• Figure 4: Different input combinations. In the first row (Example 1), both amplitude and phase are semantic information. In the second row (Example 2), the amplitude is semantic information, while the phase is sentiment polarity information.
• Figure 5: Real and imaginary part vectors after LSTM and self-attention processing.