Completion of irregular emotion sequence based on users' social relationships and historical emotions

Abstract

The use of multivariate time series is often impeded by the discontinuity of key missing features. In social networks, the absence of individual sentiment attributes presents challenges to sentiment-driven applications, such as sentiment prediction. Traditional missing value imputation fails in this field as it overlooks the interplay between missing and observed features. This paper introduces a novel deep learn-ing model that captures and assimilates the evolving features of users and their neighbors for effective sentiment completion. Experimental evidence shows that our model outperforms five other methods in performance and efficiency.