Detection of Misleading Information (DMI)

The problem of detecting intentionally misleading information in open sources requires accurate classification of the consistency of that information. Consistency can be measured internally to the information as well as by reference to background knowledge. In this effort, IET is combining natural language processing techniques and Bayesian classification to identify the fundamental characteristics of consistency for a selected domain. In this approach, assertions are extracted from each piece of source information. These assertions are compared to each other to assess their mutual consistency. Then, queries, which can be issued against open sources of background knowledge, are created for each assertion. These queries are run and their results analyzed to provide external consistency measures. The combinations of assertions, queries and answers are then used to construct a Bayesian Network that is trained to correctly classify the source information with respect to misleading content. The nodes of this network identify the characteristics that will reliably predict consistency.