[Paper Review] Principles for Developing a Knowledge Graph of Interlinked Events from News Headlines on Twitter

Saeedeh Shekarpour, Ankita Saxena+3 more|arXiv (Cornell University)|Aug 6, 2018

Semantic Web and Ontologies17 references3 citations

TL;DR

This paper presents a domain-agnostic pipeline for building a knowledge graph of interlinked events from real-time news headlines on Twitter, using a structured approach that includes event and entity annotation, interlinking across time and media, and adherence to Linked Open Data principles. The key contribution is a systematic framework based on the CEVO ontology and RDF-based event modeling, demonstrated via a Twitter news headline case study.

ABSTRACT

The ever-growing datasets published on Linked Open Data mainly contain encyclopedic information. However, there is a lack of quality structured and semantically annotated datasets extracted from unstructured real-time sources. In this paper, we present principles for developing a knowledge graph of interlinked events using the case study of news headlines published on Twitter which is a real-time and eventful source of fresh information. We represent the essential pipeline containing the required tasks ranging from choosing background data model, event annotation (i.e., event recognition and classification), entity annotation and eventually interlinking events. The state-of-the-art is limited to domain-specific scenarios for recognizing and classifying events, whereas this paper plays the role of a domain-agnostic road-map for developing a knowledge graph of interlinked events.

Motivation & Objective

• To address the lack of high-quality, semantically annotated, real-time event data in existing Linked Open Data sources.
• To develop a systematic, reusable pipeline for constructing knowledge graphs of interlinked events from unstructured, real-time text.
• To enable support for complex, time- and provenance-restricted queries on recent events, such as those involving specific locations, times, and publishers.
• To provide a standardized approach for event annotation, entity linking, and interlinking across media and temporal contexts.
• To ensure the resulting knowledge graph is consistent, dereferenceable, and interlinked with external RDF sources via LOD principles.

Proposed method

• Adopting the CEVO ontology as a background data model to represent fine-grained event types and their arguments.
• Designing a multi-stage pipeline: (1) selecting a background data model, (2) event and entity annotation via NLP techniques, (3) interlinking events across time and media using SKOS:related.
• Applying RDF serialization to publish event triples with globally unique, dereferenceable URIs.
• Ensuring data consistency through de-duplication, validation, and alignment with external knowledge bases.
• Using Twitter news headlines as a real-time data stream to demonstrate the framework’s feasibility.
• Leveraging existing NLP and information extraction techniques for event recognition, classification, and relation extraction.

Experimental results

Research questions

• RQ1 What criteria should guide the selection of a background data model for representing and interlinking events in a knowledge graph?
• RQ2 How can events and their associated entities be systematically annotated from unstructured, real-time text such as news headlines on Twitter?
• RQ3 What strategies enable effective interlinking of events across different time points and media sources?
• RQ4 How can the resulting event knowledge graph be published in compliance with Linked Open Data principles?
• RQ5 To what extent can this framework be generalized across domains beyond news headlines?

Key findings

• The proposed pipeline enables the construction of a semantically rich, event-focused knowledge graph from real-time, unstructured Twitter news headlines.
• The use of the CEVO ontology allows for fine-grained representation of event types and their arguments, supporting complex event reasoning.
• Interlinking events using SKOS:related facilitates the modeling of causal, temporal, and related event chains across time and media.
• The framework ensures compliance with Linked Open Data principles through dereferenceable URIs, data validation, and external interlinking.
• The case study demonstrates the feasibility of extracting and publishing event triples from a dynamic, real-time data stream like Twitter.
• The approach provides a scalable, domain-agnostic foundation for future systems requiring real-time, provenance-aware, and time-annotated event knowledge.