Quantum-inspired Interactive Networks for Conversational Sentiment Analysis
Published in The 28th International Joint Conference on Artificial Intelligence (IJCAI2019), 2019
Recommended citation: Yazhou Zhang, Qiuchi Li, Peng Zhang, Panpan Wang and Dawei Song (2019). "Quantum-inspired Interactive Networks for Conversational Sentiment Analysis." The 28th International Joint Conference on Artificial Intelligence (IJCAI2019). https://qiuchili.github.io/files/ijcai19-long.pdf
Conversational sentiment analysis is an emerging, yet challenging Artificial Intelligence (AI) subtask. It aims to discover the affective state of each person in a conversation. There exists a wealth of interaction information that affects the sentiment of speakers, which is crucial but largely ignored by existing sentiment analysis works. To fill this gap, we tackle the challenge of modeling intra-utterance and inter-utterance interaction dynamics for conversational sentiment analysis by inspiring from quantum theory, which has exhibited unique advantages in capturing inter-feature correlations for text modeling. We propose the quantum-inspired interactive networks (QIN), which leverages LSTM network and the mathematical formalism of quantum theory (QT), to learn both such dynamics. Specifically, a density matrix-based CNN (DM-CNN) is proposed to capture the interactions within each utterance (i.e., the correlations between words), and a strong-weak influence model inspired by quantum measurement theory is developed to learn the interactions between adjacent utterances (i.e., how one speaker influences another). Extensive experiments are conducted on the MELD dataset. The experimental results demonstrate the effectiveness of the QIN model.