Unsupervised Sentiment Analysis of Twitter Posts Using Density Matrix Representation

摘要

Nowadays, a series of pioneering studies provide the evidence that quantum probability theory can be applied in information retrieved as a mathematical framework, such as Quantum Language Model (QLM) and its variants. In these studies, the density matrix, which is defined on the quantum probabilistic space, is used to represent query and document. However, these studies are only designed for information retrieval tasks, which are unable to model sentiment information. In this paper, we investigate the feasibility of quantum probability theory for twitter sentiment analysis, and propose a density matrix based unsupervised sentiment analysis approach. The main idea is to artificially create two sentiment dictionaries, generate density matrices of documents and dictionaries using an extended QLM, then employ the quantum relative entropy to judge the similarity between density matrices of documents and dictionaries. Extensive experiments are conducted on two widely used twitter datasets, which are the Obama-McCain Debate (OMD) dataset and Sentiment Strength Twitter Dataset (SS-Tweet). The experimental results show that our approach significantly outperforms a number of baselines, demonstrating the effectiveness of the proposed density matrix based sentiment analysis approach.