Towards Social Signal Separation Based on Reconstruction Independent Component Analysis

Abstract

We all know that the ratio of social data noise is pretty significant. Therefore, tackling with noise problem is always obtained attention from data scientists. In this paper, we present a research of using reconstruction independent component analysis algorithm for blind separation social event signals from their mixtures (i.e., mixture is the combination of source signal and noise). This issue can be categorized as cocktail party problem. Despite cocktail party problem is a classical topic, however, dealing with social media data can be considered as a new research trend. From the case study with two events on Twitter, we demonstrate that our approach is quite promising. Further, our work can be applied for recommendation systems, or is used as a pre-processing step for other studies (e.g., focus search and event detection).