Multi-source Information Trustworthiness Analysis

Nowadays, a vast ocean of data from different sources is collected, and numerous applications call for the extraction of actionable insights from multi-source data. One important task is to detect untrustworthy information because such information usually indicates critical, unusual, or suspicious activities. The limitation of existing approaches is that they focus on one single source or ignore temporal information. To tackle the challenge brought by dynamic multi-source data, in this dissertation, we propose a multi-source information trustworthiness analysis framework. We represent the data as high-dimensional tensors and then apply joint tensor factorization techniques to find the common subspace across multiple sources, based on which untrustworthy information is detected. In the future, we will consider unique characteristics of various application domains and develop effective trustworthiness analysis approaches for these applications. We will also develop approaches to speed up the framework for processing large-scale data based on parallel tucker decomposition or low rank representation.