Building Journalistic Knowledge Graphs for Exploratory Text Analysis

Description

This project is a partnership between the Department of Computer Science at the University of Manchester and the BBC and it focuses on the coordination of multiple text classification and information extraction techniques for the construction of knowledge graphs (data structures which can support computers in text understanding tasks) which can support journalists in complex analytical tasks. The PhD student will work at the interface between Natural Language Processing (NLP) and Knowledge Representation, proposing and evaluating inter and intra-sentence meaning representations models to address analytical demands of journalists in the real-world. The project will explore the application of multiple state-of-the-art machine learning techniques for discourse analysis such as story/narrative extraction, argumentation mining, opinion mining.

Objectives:

- Construction of a knowledge graph-based representation using information extraction and text classification methods.
- Intrinsic evaluation of the quality of the knowledge graph extraction on the BBC corpus.
- Extrinsic evaluation using question answering and textual entailment in a journalistic setting

Research questions:

- Can the use of discourse-level and sentence-level representations (latent and explicit) over a journalistic corpus, support deeper textual inference tasks?

Novel engineering:

- The project focuses on understanding how discourse-level representations (relationships between sentences) can impact text inference tasks.