This topic is part of the FactCheck research project.
FactCheck is a framework for detecting and resolving conflicting data on the Web. It establishes an entire fact comparison process that consists of data acquisition, data comparison, the presentation of comparison results, and comprehensive analysis functions. FactCheck is a leading research topic of our research group and bears challenges in many aspects.
We define facts as pieces of information that are published by data providers (e.g., as textual content in their website(s)). If two or more websites publish data on the same topic, we humans can compare the data critically. However, this task is quite difficult for a machine, as they do not have an inherent understanding of semantics.
To execute comparisons in our FactCheck framework, we require data from various websites. Most often, structured data is not uniformly encoded, since multiple metadata encoding frameworks are available (Microdata, RDF, etc.). Since our plan is to process these data points, they need to be mapped into a common format. Furthermore, sometimes datasets use vocabularies (like schema.org) as their source for semantic information. However, we may identify multiple predicates that define the same target. Hence, we intend to reduce the diversity of predicates. Naturally, we also want to be aware of the changes and the applied updates to both the format and data points.
The goal of this project is to develop a tool to implement the structured extraction logic for FactCheck using well-established orchestration tools. Given an input file or URL, the tool should build a custom pipeline that evaluates the input against the semantic model and casts the output into a structured format. To highlight differences in performance and usability (as well as other markers), the project should be implemented as (at least) two different variants, utilizing diverse orchestration systems.
Technologies: Python; Docker; Data Pipelines; Web Services
Tags: FactCheck; Semantic Web; Structured Data; Berger
