Explainable knowledge integrated sequence model for detecting fake online reviews

Online reviews have a great influence on customers' shopping decisions. However, countless fake reviews are posted on shopping platforms, which seriously interfere with customers' shopping decisions and pollute the fair e-commerce environment. In this paper, we propose EKI-SM, an explainable knowledge integrated sequence model, to detect fake reviews. Compared with existing models, the EKI-SM displays four advantages: 1) It integrates a set of important knowledge and learns high-dimensional word embedding from reviews to guide fake review detection tasks; in addition, this knowledge explains the results of the model. 2) It learns a continuous sequence model from discrete observations with high-dimensional features, which helps to learn more discriminating fake review features. 3) It fuses the one-dimensional convolutional network, the long short-term memory network, and the residual connector to capture the local and global dependency of the sequence and make the prediction model more robust. 4) Inspired by the idea of interpretable deep learning, we explain the EKI-SM and find the important critical words for detecting fake online reviews, which derive some interesting insights. Experiments on actual fake review datasets demonstrate that the EKI-SM achieves higher accuracy in fake review detection than that of other state-of-the-art methods; indeed, it benefits from the integration of knowledge and multi-modal features.